This action might not be possible to undo. Are you sure you want to continue?

Licence Toolbox home page Discussion group

LGPL http://www.petercorke.com/vision http://groups.google.com.au/group/robotics-tool-box

Copyright c 2011 Peter Corke peter.i.corke@gmail.com September 2011 http://www.petercorke.com

Preface

This, the third release of the Toolbox, represents a decade of development. The last release was in 2005 and this version captures a large number of changes over that period but with extensive work over the last two years to support my new book “Robotics, Vision & Control” shown to the left.

Peter C0rke

Robotics,

Vision and Control

The practice of robotics and computer vision each involve the application of computational algorithms to data. The research community has developed a very large body of algorithms but for a newcomer to the ﬁeld this can be quite daunting. For more than 10 years the author has maintained two opensource matlab® Toolboxes, one for robotics and one for vision. They provide implementations of many important algorithms and allow users to work with real problems, not just trivial examples. This new book makes the fundamental algorithms of robotics, vision and control accessible to all. It weaves together theory, algorithms and examples in a narrative that covers robotics and computer vision separately and together. Using the latest versions of the Toolboxes the author shows how complex problems can be decomposed and solved using just a few simple lines of code. The topics covered are guided by real problems observed by the author over many years as a practitioner of both robotics and computer vision. It is written in a light but informative style, it is easy to read and absorb, and includes over 1000 matlab® and Simulink® examples and ﬁgures. The book is a real walk through the fundamentals of mobile robots, navigation, localization, armrobot kinematics, dynamics and joint level control, then camera models, image processing, feature extraction and multi-view geometry, and ﬁnally bringing it all together with an extensive discussion of visual servo systems.

Corke

Robotics,

Vision and Control

Peter Corke

1

Robotics, Vision and Control

isbn 978-3-642-20143-1

9 783642 201431

›

springer.com

The Machine Vision Toolbox (MVTB) provides many functions that are useful in machine vision and vision-based control. It is a somewhat eclecFUNDAMENTAL tic collection reﬂecting my personal interest in areas ALGORITHMS IN MATLAB® of photometry, photogrammetry, colorimetry. It includes over 100 functions spanning operations such 123 as image ﬁle reading and writing, acquisition, display, ﬁltering, blob, point and line feature extraction, mathematical morphology, homographies, visual Jacobians, camera calibration and color space conversion. The Toolbox, combined R with MATLAB and a modern workstation computer, is a useful and convenient environment for investigation of machine vision algorithms. For modest image sizes the processing rate can be sufﬁciently “real-time” to allow for closed-loop control. Focus of attention methods such as dynamic windowing (not provided) can be used to increase the processing rate. With input from a ﬁrewire or web camera (support provided) and output to a robot (not provided) it would be possible to implement a visual R servo system entirely in MATLAB . An image is usually treated as a rectangular array of scalar values representing intenR sity or perhaps range. The matrix is the natural datatype for MATLAB and thus makes the manipulation of images easily expressible in terms of arithmetic statements R in MATLAB language. Many image operations such as thresholding, ﬁltering and R statistics can be achieved with existing MATLAB functions. The Toolbox extends this core functionality with M-ﬁles that implement functions and classes, and mex-ﬁles for some compute intensive operations. It is possible to use mex-ﬁles to interface with image acquisition hardware ranging from simple framegrabbers to robots. Examples for ﬁrewire cameras under Linux are provided. The routines are written in a straightforward manner which allows for easy underR standing. MATLAB vectorization has been used as much as possible to improve efﬁciency, however some algorithms are not amenable to vectorization. If you have the

Machine Vision Toolbox for MATLAB

R

3

Copyright c Peter Corke 2011

This toolbox is not a clone of the Mathwork’s own Image Processing Toolbox (IPT) although there are many functions in common. though this is much less signiﬁcant today than it was in the past. This toolbox predates IPT by many years. It was developed under Unix and Linux systems and some functions rely on tools and utilities that exist only in that environment. Vision & Control” provides a detailed discussion (over 600 pages.MATLAB compiler available then this can be used to compile bottleneck functions. contains many functions that are useful for image feature extraction and control. and I commend it to you. R R Machine Vision Toolbox for MATLAB R 4 Copyright c Peter Corke 2011 . This toolbox considers images generally as arrays of double precision numbers. This is extravagant on storage. nearly 400 ﬁgures and 1000 code examples) of how to use the Toolbox functions to solve many types of problems in robotics. is open-source. The manual is now auto-generated from the comments in the MATLAB code itself which reduces the effort in maintaining code and a separate manual as I used to — the downside is that there are no worked examples and ﬁgures in the manual. Some particularly compute intensive functions are provided as mex-ﬁles and may need to be compiled for the particular platform. However the book “Robotics.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . AxisWebCamera . . . . . . . . . . . . 2 . . . . . . . . . .6 Use in research . . . . . . . . . . Movie . . . . . . . . . . . . . . . . . . . . . CatadioptricCamera . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . SphericalCamera . . . . . . . . . . . . . . . . . . . about . . . . . . . . . . . . . . . .5 Use in teaching . . . . . . . . . . . . 1 Introduction 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PointFeature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ScalePointFeature . . . . . . . . . . . . . . . . . . . . . . . angdiff . . . . . . . . . . . . RegionFeature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FeatureMatch . . . . blackbody . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Contents Introduction . . . . . . .3 How to obtain the Toolbox 1. . . . . . . . . . . . . . . . . . . . . PGraph . . . . . BagOfWords . . . . Polygon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hough . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . CentralCamera . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ray3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . .1 Other toolboxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Support . . . . Video . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 MATLAB version issues . . . Tracker . . . . SiftPointFeature . . . . . . . . . . . . . . . . . . LineFeature . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . Functions and classes Camera . . . . .1 What’s new . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . 3 10 10 10 11 11 11 11 12 12 13 13 19 29 32 37 40 41 45 48 53 55 59 62 63 70 73 78 80 83 85 87 88 89 89 90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6. . . . . . . . . SurfPointFeature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Machine Vision Toolbox for MATLAB R 5 Copyright c Peter Corke 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . anaglyph . . . . . FishEyeCamera . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . colordistance colorize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . igraphseg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ccdresponse . . . . . . . . . . . . . . . . . . . . . . . . . distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . icp . . . . . . . . . . . . . . . . . . . . . . . 90 90 91 91 92 92 92 93 94 94 95 95 96 96 97 98 99 99 100 100 101 101 102 102 103 103 103 104 104 105 106 106 107 107 108 109 110 110 111 112 113 114 115 116 116 117 117 118 119 120 Machine Vision Toolbox for MATLAB Copyright c Peter Corke 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . icolor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . epidist . . . . . . . . . . . . colorkmeans . . . . . . . . . . . . . . . . . . . . colorname . . idouble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . icanny . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iconv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iclose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . cie primaries circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . humoments . . . . . . . . . . . . . . . . . . . . . . . . . . . homwarp . . . . . . . . . . . . . . . colorspace . . . . . . . . . . . colnorm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 . . . . . . . . . . igamma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gauss2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . homography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . fmatrix . . . . . . . . . . . . . . . . . . . epiline . . . . . . . . . . . . . . . . . . . diff2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iconcat . . . . . . . . cmfrgb . . . . . . . . . . . . . . . . . . . . . homtrans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ihist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . homline . . . . . . . . . . . . .CONTENTS CONTENTS boundmatch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . edgelist . . . idecimate . . . . . . . . . . . . . . . . col2im . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iblobs . . . . . . . . . . . . . . . . . . . . iendpoint . . . . . . . . . . . . . . . . ibbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . idisp . . . . . hitormiss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . cmfxyz . . . . . . . . . . . . . . . . . . . . . . . . . bresenham . . . . . . . . . . icorner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . e2h . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ianimate . . . . . . . . . . . . . . . . . . . . . closest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iint . . . . . . . . . . . . . . . . . camcald . . . h2e . idisplabel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R . . . irectify . . . . . irank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . isamesize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ireplicate . . . im2col . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ithresh . . . kcircle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ipad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . imorph . . . . . . . . . . . . . . . . . . . . . . . 120 120 121 122 122 123 123 124 125 125 126 126 127 127 128 128 129 129 130 131 132 132 133 133 134 134 135 135 136 136 136 137 138 138 139 140 140 141 142 143 143 144 144 145 145 146 147 147 148 148 Machine Vision Toolbox for MATLAB Copyright c Peter Corke 2011 . . . irotate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ipyramid . . . . . . . . . . . . . . . . . . . . imatch . . . . . . . . . . . . . . . . . . . isize . . . . . . . . . . . . . . iwindow . . . . . . . . . . . . . . . . . isrot . . . . . . . . . . . . . . . . . . . . . . . . . kdog . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iscalemax . isimilarity . . . . . . . . .CONTENTS CONTENTS iisum . . . . . . . . . ismooth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . imeshgrid . . . . . . . . . . . . . . . . . . . . . ivar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iscalespace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 . . . . . . . . itriplepoint . . . . . . . . . . . . . . . . . . iproﬁle . . . . . . . . . . . . isobel . . . . . . . . . . . . . . . . ilabel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . inormhist . . . . . . . . . . . . . . . . . . . . . . . iscolor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . isift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . imono . . . . . . . . . . . . iroi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . imoments . . . . . . . . . . . . . . . . . . . . . . . isvec . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . kdgauss . . . itrim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . imser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iopen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ipaste . . . . . . . . . . . . . . . . . . . . . . . . . . . . . istereo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ishomog . . . ithin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . istretch . . . . . . . . . iscale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . intgimage . . . isurf . . . . . . . . . . . . . . . . . . ipixswitch . . . . . . . . . . . . . . kgauss .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ramp . . . . . . . . . rotx . . . . . . mlabel . . . . . rotz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . radgrad . . . . . . . . . . . . . . . . . . . . ransac . . . . . . . . . . . . . . . . . . . . . . . . . niblack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . lambda2xy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . plotp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . lambda2rg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . rg addticks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . luminos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . peak . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . kmeans . . . . . . . . . . plot ellipse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . peak2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mtools . . . . . . . . . . . . . . . . . . . . .CONTENTS CONTENTS klaplace . . . . plot homline . . . . . . . . . . . . . . . . . roty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . numcols . . . . . r2t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . numrows . . . . . . . . . plot sphere . . . plot poly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . medﬁlt1 . . . . . . . . . . . . . . . . . . . . pnmﬁlt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . pgmﬁlt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ksobel . . . . . . . . . mpq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . plot2 . . . . . . . . . . . . . . . . . . . . . . . otsu . . . . . . . . . . . . . . . . . . . norm2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . rluminos . . . . . . . . . . . 149 149 150 150 151 151 152 152 153 153 153 154 154 155 155 155 156 156 156 157 158 158 158 159 159 160 160 161 162 162 163 163 164 164 164 165 165 166 166 167 167 168 168 169 169 171 171 172 172 172 Machine Vision Toolbox for MATLAB Copyright c Peter Corke 2011 . . . . . . . . . . ktriangle . . . . . . . . . . . . . . . . . . . . . . . . . . . plot circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . mkcube . . . . . . . . . . . . . . . . . . . . plot point . . . . . loadspectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . maxﬁlt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq poly . . . . . . . . 8 . . . . . ncc . . . . . . . . . . . . . . . . . . mkgrid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . plot box . . . . . . . . . . . . npq poly . . . . . . . . . . . . . . . . npq . mplot . . . . . . . . . . . . . . . . . . . . . . . . . . . . klog . . . . . . . . plot ellipse inv plot frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xaxis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . trotz . . . . t2r . . . . . . . . . . . . . . . . . . . . . . . . tr2angvec . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . trnorm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . upq . . . . . . . . . . . . trprint . transl . . . . . . . testpattern . . . . . . . . . . . . . . . . . . se2 . . . . . . . . . . . . tr2rt . . . . . . . . . .CONTENTS CONTENTS rpy2tr . . . . . . . . . . . . . . . . . . rt2tr . . . . . . . zsad . . . . . . . . . . . . . . . . . . . . . . . . . . troty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . stdisp . . . . . . . . . . . tristim2cc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . unit . . . . . . . . . . . . . . . . . . . . . . . . . . tpoly . . . . . . . . . . zcross . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xycolorspace yaxis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ssd . . . . . . . . . . . . . . . . . . . . . . . . . . . . useﬁg . . . . . . . . . . . . . . . . 173 173 174 174 175 175 175 176 176 177 178 179 179 180 180 181 181 182 182 183 183 184 184 185 185 185 186 186 187 187 188 188 189 Machine Vision Toolbox for MATLAB R 9 Copyright c Peter Corke 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vex . . . . . . . sad . . . . . . . . . . . . . . . . . . . . trotx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . zncc . . . . . . . . . . . . . . . tr2rpy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . tb optparse . . . . . . . . . . . . . . . . . . . . . skew . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . upq poly . . . . . . . . . . . . . zssd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

That’s what you your teachers. and you will be suitably acknowledged.google. I am very happy to accept contributions for inclusion in future versions of the toolbox.2 Support There is no support! This software is made freely available in the hope that you ﬁnd it useful in solving whatever problems you have to hand. I can guarantee that I will not respond to any requests for help with assignments or homework. You need to signup in order to post.au/group/robotics-tool-box which is a forum for discussion. tutors.Chapter 1 Introduction 1.com. and the signup process is moderated by me so allow a few days for this to happen. I need you to write a Machine Vision Toolbox for MATLAB R 10 Copyright c Peter Corke 2011 . lecturers and professors are paid to do. I am happy to correspond with people who have found genuine bugs or deﬁciencies but my response time can be long and I can’t guarantee that I respond to your email. You might instead like to communicate with other users via the Google Group called “Robotics Toolbox” http://groups. no matter how urgent or important they might be to you.1 • x New features: • x Bugﬁxes: • Improved error messages in many functions • Removed trailing commas from if and for statements What’s new Changes: 1.

A menu-driven demonstration can be invoked by the function rtdemo. Corke}. If you plan to distribute paper copies of the PDF manual then every copy must include the ﬁrst two pages (cover and licence). Machine Vision Toolbox for MATLAB R 11 Copyright c Peter Corke 2011 . 1. It is R auto-generated from the comments in the MATLAB code and is fully hyperlinked: to external web sites. This is just a means for me to gauge interest and to help convince my bosses (and myself) that this is a worthwhile activity. Volume = {12}.pdf or the web-based documentation html/*. The details are @article{Corke05f. type of organization and application.pdf is a manual that describes all functions in the Toolbox.I. The web page requests some information from you regarding such as your country. Journal = {IEEE Robotics and Automation Magazine}.zip).3.html on a server for class use. HOW TO OBTAIN THE TOOLBOX CHAPTER 1. Month = nov. INTRODUCTION few words about why you want to join the list so I can distinguish you from a spammer or a web-bot. 1.gz) or zip format (. 1. Year = {2005}.3 How to obtain the Toolbox The Machine Vision Toolbox is freely available from the Toolbox home page at http://www.5 Use in teaching This is deﬁnitely encouraged! You are free to put the PDF manual (vision.petercorke. and the “See also” functions to each other. Author = {P. Title = {Machine Vision Toolbox}. the table of content to functions. The ﬁle robot. Number = {4}.com The ﬁles are available in either gzipped tar format (.1. 1.4 MATLAB version issues The Toolbox has been tested under R2011a.6 Use in research If the Toolbox helps you in your endeavours then I’d appreciate you citing the Toolbox when you publish.

VLFeat http://www. graph-based segmentation and pose estimation are based on great code Some of the MEX ﬁle use some really neat macros that were part of the package VISTA Copyright 1993. RANSAC code by Peter Kovesi. See the ﬁle CONTRIB for details. pp 16–25.7 Acknowledgements Last. Twente.I. Corke. P. MSER. INTRODUCTION Pages = {16-25} } or “Machine Vision Toolbox”. and there are hundreds of modules available.6. IEEE Robotics and Automation Magazine. pose estimation by Francesco Moreno-Noguer.7.Functions such as SURF.1. Coimbra. 1994 University of British Columbia. contributed by Nuno Alexandre Cid Martins of I.org is a great collection of advanced computer vision algorithms for MATLAB.mathworks. numerical routines for geometric vision by various members of the Visual Geometry Group at Oxford (from the web site of the Hartley and Zisserman book. this release includes functions for computing image plane homographies and the fundamental matrix. ACKNOWLEDGEMENTS CHAPTER 1. Vincent Lepetit.com/matlabcentral is a great resource for user contributed MATLAB code. and the SURF feature detector by Dirk-Jan Kroon at U.vlfeat. Pascal Fua at the CVLab-EPFL. November 2005. 1. 1.R. but not least. which is also given in electronic form in the CITATION ﬁle. color space conversions by Pascal Getreuer.1 Other toolboxes Matlab Central http://www.the graph-based image segmentation software by Pedro Felzenszwalb.. The Camera Calibration Toolbox by Jean-Yves Bouguet is used unmodiﬁed. the k-means and MSER algorithms by Andrea Vedaldi and Brian Fulkerson.S. 12(4). Machine Vision Toolbox for MATLAB R 12 Copyright c Peter Corke 2011 .

Chapter 2 Functions and classes Camera Camera superclass An abstract superclass for Toolbox camera classes. Methods plot hold ishold clf ﬁgure mesh point line plot camera rpy move centre delete char display plot projection of world point to image plane control ﬁgure hold for image plane window test ﬁgure hold for image plane clear image plane ﬁgure holding the image plane draw shape represented as a mesh draw homogeneous points on image plane draw homogeneous lines on image plane draw camera in world view set camera attitude clone Camera after motion get world coordinate of camera centre object destructor convert camera parameters to string display camera parameters Properties (read/write) npix pp rho T image dimensions (2 × 1) principal point (2 × 1) pixel dimensions (2 × 1) in metres camera pose as homogeneous transformation R Machine Vision Toolbox for MATLAB 13 Copyright c Peter Corke 2011 .

• The object can create a window to display the Camera image plane. P ‘pixel’.CHAPTER 2. FUNCTIONS AND CLASSES Properties (read only) nu nv u0 v0 number of pixels in u-direction number of pixels in v-direction principal point u-coordinate principal point v-coordinate Notes • Camera is a reference object. T ‘color’. C Name of camera Load image IM to image plane Image plane resolution: N × N or N=[W H] Image sensor size in metres (2 × 1) [metres] Principal point (2 × 1) Pixel size: S × S or S=[W H] Standard deviation of additive Gaussian noise added to returned image projections Pose of the camera as a homogeneous transformation Color of image plane background (default [1 1 0. C = Camera(options) creates a default (abstract) camera with null parameters.8]) Notes • Normally the class plots points and lines into a set of axes that represent the image plane. used by all subclasses. IM ‘resolution’. Options ‘name’. SIGMA ‘pose’. S ‘centre’. Camera. S ‘noise’. N ‘image’. The ‘image’ option paints the speciﬁed image onto the image plane and allows points and lines to be overlaid. this window is protected and can only be accessed by the plot methods of this object. • Camera objects can be used in vectors and arrays • This is an abstract class and must be subclassed and a project() method deﬁned. Machine Vision Toolbox for MATLAB R 14 Copyright c Peter Corke 2011 . N ‘sensor’.Camera Create camera object Constructor for abstact Camera class.

centre() is the 3-dimensional position of the camera centre (3 × 1). Camera. ﬁsheyecamera.char() is a compact string representation of the camera parameters.delete Camera object destructor C.delete() destroys all ﬁgures associated with the Camera object and removes the object.clf Clear the image plane C.char Convert to string s = C.display() displays a compact human-readable representation of the camera parameters. Camera. Camera.centre Get camera position p = C. Machine Vision Toolbox for MATLAB R 15 Copyright c Peter Corke 2011 . CatadioptricCamera.CHAPTER 2. Camera. SphericalCamera Camera.clf() removes all graphics from the camera’s image plane.display Display value C. FUNCTIONS AND CLASSES See also CentralCamera.

char Camera.hold() sets “hold on” for the camera’s image plane.CHAPTER 2.hold Control hold on image plane graphics C. Camera.ishold() returns true (1) if the camera’s image plane is in hold mode. otherwise false (0).line(L) plots lines on the camera image plane which are deﬁned by columns of L (3 × N ) considered as lines in homogeneous form: a.ﬁgure Return ﬁgure handle H = C.v + c = 0.line Plot homogeneous lines on image plane C.ﬁgure() is the handle of the ﬁgure that contains the camera’s image plane graphics. and off if H is false (or 0). Camera.hold(H) hold mode is set on if H is true (or > 0). Camera. C. Machine Vision Toolbox for MATLAB R 16 Copyright c Peter Corke 2011 .u + b.ishold Return image plane hold status H = C. FUNCTIONS AND CLASSES Notes • This method is invoked implicitly at the command line when the result of an expression is a Camera object and the command has no trailing semicolon. See also Camera.

mesh Plot mesh object on image plane C.T.hold. Camera. Temporarily overrides the current camera pose C. y.move(T) is a new camera object that is a clone of C but its pose is displaced by the homogeneous transformation T with respect to the current pose of C. FUNCTIONS AND CLASSES Camera. y. See also mesh. If p is 2 × N the points are assumed to be image plane coordinates and are plotted directly. options) projects world points p (3 × N ) to the image plane and plots them.plot Plot points on image plane C. Camera. sphere.plot. T Transform all points by the homogeneous transformation T before projecting them to the camera image plane. uv = C. Set the camera pose to the homogeneous transformation T before projecting points to the camera image plane.plot(p. options) projects a 3D shape deﬁned by the matrices x.clf Camera. z to the image plane and plots them. mkcube.plot(p) as above but returns the image plane coordinates uv (2 × N ). The matrices x. z are of the same size and the corresponding elements of the matrices deﬁne 3D points.CHAPTER 2.mesh(x. Camera. Options ‘Tobj’. T ‘Tcam’.move Instantiate displaced camera C2 = C. Machine Vision Toolbox for MATLAB R 17 Copyright c Peter Corke 2011 . If p has 3 dimensions (3xNxS) then it is considered a sequence of point sets and is displayed as an animation. cylinder. z. Camera. y.

point(p) plots points on the camera image plane which are deﬁned by columns of p (3 × N ) considered as points in homogeneous form. Overrides the current camera pose C.plot camera(options) draw a camera as a simple 3D model in the current ﬁgure.T. Machine Vision Toolbox for MATLAB R 18 Copyright c Peter Corke 2011 . ‘fps’.point Plot homogeneous points on image plane C.2 x maximum axis dimension) Notes • The graphic handles are stored within the Camera object.CHAPTER 2. S Text size for annotation (default 12) ‘drawnow’ Execute MATLAB drawnow function Additional options are considered MATLAB linestyle parameters and are passed directly to plot.hold.mesh. T Set the camera pose to the homogeneous transformation T before projecting points to the camera image plane. Camera. N Number of frames per second for point sequence display ‘sequence’ Annotate the points with their index ‘textcolor’. C Text color for annotation (default black) ‘textsize’. ‘Tcam’. ‘Tobj’. Camera. FUNCTIONS AND CLASSES Options Transform all points by the homogeneous transformation T before projecting them to the camera image plane. T See also Camera. S Camera displayed in pose T (homogeneous transformation 4 × 4) Overall scale factor (default 0. Camera.clf Camera. T ‘scale’. Options ‘Tcam’.plot camera Display camera icon in world view C.

Machine Vision Toolbox for MATLAB R 19 Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES Camera. The image is not inverted.rpy(rpy) as above but rpy=[R. y) sets the camera attitude to the speciﬁed roll-pitch-yaw angles.y]. The camera coordinate system is: 0------------> u X | | | + (principal point) | | Z-axis is into the page.rpy(R. a subclass of Camera.rpy Set camera attitude C. p. v Y This camera model assumes central projection.p. CentralCamera Perspective camera class A concrete class for a central-projection perspective camera. C. that is.CHAPTER 2. the focal point is at z=0 and the image plane is at z=f.

CHAPTER 2. FUNCTIONS AND CLASSES Methods project K C H invH F E invE fov ray plot hold ishold clf ﬁgure mesh point line plot camera plot line tr plot epiline ﬂowﬁeld visjac p visjac p polar visjac l visjac e rpy move centre estpose delete char display project world points camera intrinsic matrix camera matrix camera motion to homography decompose homography camera motion to fundamental matrix camera motion to essential matrix decompose essential matrix ﬁeld of view Ray3D corresponding to point plot projection of world point on image plane control hold for image plane test ﬁgure hold for image plane clear image plane ﬁgure holding the image plane draw shape represented as a mesh draw homogeneous points on image plane draw homogeneous lines on image plane draw camera in world view draw line in theta/rho format draw epipolar line compute optical ﬂow image Jacobian for point features image Jacobian for point features in polar coordinates image Jacobian for line features image Jacobian for ellipse features set camera attitude clone Camera after motion get world coordinate of camera centre estimate pose object destructor convert camera parameters to string display camera parameters Properties (read/write) npix pp rho f k p distortion T image dimensions in pixels (2 × 1) intrinsic: principal point (2 × 1) intrinsic: pixel dimensions (2 × 1) in metres intrinsic: focal length intrinsic: radial distortion vector intrinsic: tangential distortion parameters intrinsic: camera distortion [k1 k2 k3 p1 p2] extrinsic: camera pose as homogeneous transformation Machine Vision Toolbox for MATLAB R 20 Copyright c Peter Corke 2011 .

C Camera matrix C = C.CHAPTER 2. CentralCamera. • Camera objects can be used in vectors and arrays See also Camera CentralCamera. Machine Vision Toolbox for MATLAB R 21 Copyright c Peter Corke 2011 .C() is the 3×4 camera matrix. C = CentralCamera(options) as above but with speciﬁed parameters. also known as the camera calibration or projection matrix. FUNCTIONS AND CLASSES Properties (read only) nu nv u0 v0 number of pixels in u-direction number of pixels in v-direction principal point u-coordinate principal point v-coordinate Notes • Camera is a reference object.CentralCamera Create central projection camera object C = CentralCamera() creates a central projection camera with canonic parameters: f=1 and name=’canonic’.

CatadioptricCamera. ﬁsheyecamera. S. Display an image rather than points Image plane resolution: N × N or N=[W H] Image sensor size in metres (2 × 1) Principal point (2 × 1) Pixel size: S × S or S=[W H] Standard deviation of additive Gaussian noise added to returned image projections Pose of the camera as a homogeneous transformation Color of image plane background (default [1 1 0. S ‘centre’. IM ‘resolution’.Kosecka. E = C.Ma.and v-axes parallel to x. C Name of camera Focal length [metres] Distortion vector [k1 k2 k3 p1 p2] Distortion vector [k1 k2 p1 p2 k3] Default camera parameters: 1024 × 1024.8]) See also Camera.CHAPTER 2. Springer. “An invitation to 3D”. f=8mm.E Essential matrix E = C. camera at origin. N ‘sensor’. D ‘default’ ‘image’.and y-axes respectively.E(T) is the essential matrix relating two camera views. CentralCamera. The ﬁrst view is from the current camera pose C. 2003. S. S ‘noise’. p. N ‘focal’.F. FUNCTIONS AND CLASSES Options ‘name’. SphericalCamera CentralCamera.F(F) is the essential matrix based on the fundamental matrix F (3 × 3) and the intrinsic parameters of camera C. optical axis is z-axis. E = C.T and the second is a relative motion represented by the homogeneous transformation T. J. 10um pixels.177 See also CentralCamera.invE Machine Vision Toolbox for MATLAB R 22 Copyright c Peter Corke 2011 .Sastry. P ‘pixel’. Reference Y. u.Soatto.F(C2) is the essential matrix relating two camera views described by camera objects C (ﬁrst view) and C2 (second view). SIGMA ‘pose’. D ‘distortion-bouguet’. T ‘color’. F ‘distortion’.

The ﬁrst view is from the current camera pose C.K() is the 3 × 3 intrinsic parameter matrix. F = C. “An invitation to 3D”. d) is a 3 × 3 homography matrix for the camera observing the plane with normal n and at distance d. Reference Y. p. FUNCTIONS AND CLASSES CentralCamera.Kosecka.F(C2) is the fundamental matrix relating two camera views described by camera objects C (ﬁrst view) and C2 (second view).H Homography matrix H = C. Springer. n.Sastry.Ma. See also CentralCamera. from two viewpoints.F Fundamental matrix F = C. The ﬁrst view is from the current camera pose C.E CentralCamera.177 See also CentralCamera. S.K Intrinsic parameter matrix K = C.H(T. 2003.T and the second is a relative motion represented by the homogeneous transformation T. S.Soatto.CHAPTER 2. Machine Vision Toolbox for MATLAB R 23 Copyright c Peter Corke 2011 .F(T) is the fundamental matrix relating two camera views. J.H CentralCamera.T and the second is after a relative motion represented by the homogeneous transformation T.

Machine Vision Toolbox for MATLAB R 24 Copyright c Peter Corke 2011 .ﬂowﬁeld(v) displays the optical ﬂow pattern for a sparse grid of points when the camera has a spatial velocity v (6 × 1). uv) is an estimate of the pose of the object deﬁned by coordinates xyz (3×N ) in its own coordinate frame. uv (2×N ) are the corresponding image plane coordinates. Journal on Computer Vision.estpose Estimate pose from object model and camera view T = C.CHAPTER 2. Lepetit. Reference “EPnP: An accurate O(n) solution to the PnP problem”. 2009. Feb.fov Camera ﬁeld-of-view angles. pp. a = C. Fua. CentralCamera. See also quiver CentralCamera.invE(E) decomposes the essential matrix E (3 × 3) into the camera motion. In practice there are multiple solutions and s (4x4xN) is a set of homogeneous transformations representing possible camera motion.invE Decompose essential matrix s = C. CentralCamera. 81. vol. Int. and P.fov() are the ﬁeld of view angles (2 × 1) in radians for the camera x and y (horizontal and vertical) directions. 155-166. F. V. FUNCTIONS AND CLASSES CentralCamera.estpose(xyz. Moreno-Noguer.ﬂowﬁeld Optical ﬂow C.

CHAPTER 2.Sastry. Reference Y. p116. In practice there are multiple solutions and s is a vector of structures with elements: • T.Kosecka. FUNCTIONS AND CLASSES s = C. Chap 9.3 Machine Vision Toolbox for MATLAB R 25 Copyright c Peter Corke 2011 . Reference Hartley & Zisserman. s.Ma.Soatto.Sastry.invE(E. J. 2003.invH(H) decomposes the homography H (3 × 3) into the camera motion and the normal to the plane. “An invitation to 3D”.invH Decompose homography matrix s = C.E CentralCamera. “An invitation to 3D”. translation not to scale • n. Springer. J.Kosecka. See also CentralCamera. section 5. “Multiview Geometry”. p. s. camera motion as a homogeneous transform matrix (4 × 4). s. p120122 Notes • The transformation is from view 1 to view 2.Soatto. Springer. 2003. 259 Y. s. p) as above but only solutions in which the world point p is visible are returned. normal vector to the plane (3 × 3) Notes • There are up to 4 solutions • Only those solutions that obey the positive depth constraint are returned • The required camera intrinsics are taken from the camera object • The transformation is from view 1 to view 2.Ma.

‘Tobj’.project Project world points to image plane uv = C.plot epiline(f. options) are the image plane coordinates (2 × N ) corresponding to the world points p (3 × N ). p) plots the epipolar lines due to the fundamental matrix f and the image points p.H CentralCamera.plot line tr(L) plots lines on the camera’s image plane that are described by columns of L with rows theta and rho respectively.plot epiline(f.CHAPTER 2. See also Hough CentralCamera. T Machine Vision Toolbox for MATLAB R 26 Copyright c Peter Corke 2011 . If Tcam (4x4xS) is a transform sequence then uv (2xNxS) represents the sequence of projected points as the camera moves in the world.T.plot epiline Plot epipolar line C.project(p. H = C. Options Transform all points by the homogeneous transformation T before projecting them to the camera image plane. p) as above but return a vector of graphic handles. C. ‘Tcam’. ls) as above but draw lines using the line style arguments ls. FUNCTIONS AND CLASSES See also CentralCamera. p. Temporarily overrides the current camera pose C. T Set the camera pose to the homogeneous transformation T before projecting points to the camera image plane.plot epiline(f.plot line tr Plot line in theta-rho format CentralCamera. one per line. CentralCamera.

Rives. pp.ray(p) returns a vector of Ray3D objects.ray 3D ray for image point R = C. IEEE Transactions on Robotics and Automation.b. Reference Hartley & Zisserman. “A New Approach to Visual Servoing in Robotics”. Reference B. vol. one for each point deﬁned by the columns of p. June 1992.visjac e Visual motion Jacobian for point feature J = C. The ellipse lies in the world plane pl = (a. Machine Vision Toolbox for MATLAB R 27 Copyright c Peter Corke 2011 . The Jacobian gives the rates of change of the ellipse parameters in terms of camera spatial velocity.2E2uv + 2E3u + 2E4v + E5 = 0. Espiau. F. pl) is the image Jacobian (5 × 6) for the ellipse E (5 × 1) described by u2 + E1v2 . 313-326.plot CentralCamera. and P.c. p 162 See also Ray3D CentralCamera. “Multiview Geometry”.CHAPTER 2. 8. FUNCTIONS AND CLASSES If Tobj (4x4x) is a transform sequence then uv (2xNxS) represents the sequence of projected points as the object moves in the world. Chaumette.d) such that aX + bY + cZ + d = 0. See also Camera.visjac e(E.

visjac l(L.visjac l Visual motion Jacobian for line feature J = C.c.visjac p.visjac p polar. Hutchinson.visjac l CentralCamera. FUNCTIONS AND CLASSES See also CentralCamera. Vol 12(5).visjac p(uv. Machine Vision Toolbox for MATLAB R 28 Copyright c Peter Corke 2011 . Oct. z) is the image Jacobian (2N × 6) for the image plane points uv (2 × N ). The Jacobian gives the image-plane point velocity in terms of camera spatial velocity. Hager & Corke. The Jacobian gives the rates of change of the line parameters in terms of camera spatial velocity. pl) is the image Jacobian (2N × 6) for the image plane lines L (2 × N ). 313-326. Each column of L is a line in theta-rho format. CentralCamera. 8. CentralCamera. R&A. Chaumette.CHAPTER 2. CentralCamera.visjac p Visual motion Jacobian for point feature J = C. IEEE Transactions on Robotics and Automation. or a vector (N × 1) of depth for each point.visjac p polar.visjac e CentralCamera. The depth of the points from the camera is given by z which is a scalar for all points. Rives. June 1992. and the rows are theta and rho respectively. and P. pp. Reference B. IEEE Trans. See also CentralCamera. The lines all lie in the plane pl = (a. 1996. CentralCamera.b. vol. F. Espiau.visjac p.d) such that aX + bY + cZ + d = 0. Reference “A tutorial on Visual Servo Control”. pp 651-670. “A New Approach to Visual Servoing in Robotics”.

in Proc.visjac p. CentralCamera. Machine Vision Toolbox for MATLAB R 29 Copyright c Peter Corke 2011 .visjac e CentralCamera. CentralCamera. I.visjac p polar. 2009. or a vector (N × 1) of depths for each point. Spindler. Oct. F.visjac l. The depth of the points from the camera is given by z which is a scalar for all point. CentralCamera. z) is the image Jacobian (2N × 6) for the image plane points rt (2 × N ) described in polar form. and F. radius and theta. Chaumette. The Jacobian gives the image-plane polar point coordinate velocity in terms of camera spatial velocity.visjac p polar(rt. Corke. P. Louis).visjac p polar Visual motion Jacobian for point feature J = C. CentralCamera. Reference “Combining Cartesian and polar coordinates in IBVS”. 5962-5967.visjac e SiftPointFeature SIFT point corner feature object A subclass of PointFeature for SIFT features.visjac l. pp. FUNCTIONS AND CLASSES See also CentralCamera. See also CentralCamera. Conf on Intelligent Robots and Systems (IROS). Int.CHAPTER 2. (St.

Lowe.SIFT. vol. Int.CHAPTER 2. • SiftCornerFeature objects can be used in vectors and arrays • The SIFT algorithm is patented and not distributed with this toolbox.91-110. See also isift. D. SurfPointFeature Machine Vision Toolbox for MATLAB R 30 Copyright c Peter Corke 2011 . See README. pp. Journal on Computer Vision. References “Distinctive image features from scale-invariant keypoints”.u is a 2×N matrix with each column the corresponding u coordinate. PointFeature. Nov.60. FUNCTIONS AND CLASSES Methods plot plot scale distance match ncc uv display char Plot feature position Plot feature scale Descriptor distance Match features Descriptor similarity Return feature coordinate Display value Convert value to string Properties u horizontal coordinate v vertical coordinate strength feature strength theta feature orientation [rad] scale feature scale descriptor feature descriptor (vector) index of image containing feature image id Properties of a vector of SiftCornerFeature objects are returned as a vector. You can download a SIFT implementation which this class can utilize. If F is a vector (N ×1) of SiftCornerFeature objects then F. 2004. ScalePointFeature. Notes • SiftCornerFeature is a reference object.

C ‘alpha’.match(f2. F. f = PointFeature(u. If F is a vector then each element is plotted.plot scale Plot feature scale F.CHAPTER 2. FUNCTIONS AND CLASSES SiftPointFeature. See also isift SiftPointFeature. options) is a vector of FeatureMatch objects that describe candidate matches between the two vectors of SIFT features F and f2. ls) as above but the optional line style arguments ls are passed to plot. 0=transparent (default 0. Options ‘circle’ ‘clock’ ‘arrow’ ‘disk’ ‘color’. f = PointFeature(u.match Match SIFT point features m = F. Correspondence is based on descriptor similarity.2) Machine Vision Toolbox for MATLAB R 31 Copyright c Peter Corke 2011 .plot scale(options) overlay a marker to indicate feature point position and scale.SiftPointFeature Create a SIFT point feature object f = SiftPointFeature() is a point feature object with null parameters. SiftPointFeature.plot scale(options. v. A Indicate scale by a circle (default) Indicate scale by circle with one radial line for orientation Indicate scale and orientation by an arrow Indicate scale by a translucent disk Color of circle or disk (default green) Transparency of disk. 1=opaque. strength) as above but with speciﬁed strength. v) is a point feature object with speciﬁed coordinates.

support Support region of feature out = F. The support region is scaled to w × w and rotated so that the feature’s orientation axis is upward. out = F.T] = F.support(images.CHAPTER 2. Machine Vision Toolbox for MATLAB R 32 Copyright c Peter Corke 2011 . extracted from the image im in which the feature appears. w) is an image of the support region of the feature F. w) as above but the support region is displayed.support(images. See also SiftPointFeature SphericalCamera Spherical camera class A concrete class a spherical-projection camera. w) as above but returns the pose of the feature as a 3 × 3 homogeneous transform in SE(2) that comprises the feature position and orientation. w) as above but if the features were extracted from an image sequence images then the feature is extracted from the appropriate image in the same sequence.support(im.support(im. [out. F. FUNCTIONS AND CLASSES SiftPointFeature.

FUNCTIONS AND CLASSES Methods project plot hold ishold clf ﬁgure mesh point line plot camera rpy move centre delete char display project world points plot/return world point on image plane control hold for image plane test ﬁgure hold for image plane clear image plane ﬁgure holding the image plane draw shape represented as a mesh draw homogeneous points on image plane draw homogeneous lines on image plane draw camera set camera attitude copy of Camera after motion get world coordinate of camera centre object destructor convert camera parameters to string display camera parameters Properties (read/write) npix pp rho T image dimensions in pixels (2 × 1) intrinsic: principal point (2 × 1) intrinsic: pixel dimensions (2 × 1) in metres extrinsic: camera pose as homogeneous transformation Properties (read only) nu nv number of pixels in u-direction number of pixels in v-direction Note • SphericalCamera is a reference object.CHAPTER 2. • SphericalCamera objects can be used in vectors and arrays See also Camera Machine Vision Toolbox for MATLAB R 33 Copyright c Peter Corke 2011 .

SphericalCamera Create spherical projection camera object C = SphericalCamera() creates a spherical projection camera with canonic parameters: f=1 and name=’canonic’. options) are the image plane coordinates for the world points p. The columns of p (3 × N ) are the world points and the columns of pt (2 × N ) are the corresponding spherical projection points.CHAPTER 2. N ‘pixel’. T Name of camera Pixel size: S × S or S(1)xS(2) Pose of the camera as a homogeneous transformation See also Camera. CentralCamera. CatadioptricCamera SphericalCamera. C = CentralCamera(options) as above but with speciﬁed parameters. ﬁsheyecamera.project(p. T Transform all points by the homogeneous transformation T before projecting them to the camera image plane.T. each column is phi (longitude) and theta (colatitude).project Project world points to image plane pt = C. Options ‘name’. See also SphericalCamera. T ‘Tcam’. Options ‘Tobj’. FUNCTIONS AND CLASSES SphericalCamera.plot Machine Vision Toolbox for MATLAB R 34 Copyright c Peter Corke 2011 . Set the camera pose to the homogeneous transformation T before projecting points to the camera image plane. S ‘pose’. Overrides the current camera pose C.

center of the target in world coords (0. defaults in parentheses: target size . The external view. Two windows are shown and animated: 1. FUNCTIONS AND CLASSES SphericalCamera.a stopping criteria on feature error norm (0) . showing the desired view (*) and the current view (o) Machine Vision Toolbox for MATLAB R 35 Copyright c Peter Corke 2011 .the number of iterations to run the simulation (500) . The camera/robot is initially at pose T and is driven to the orgin.the side length of the target in world units (0. params) Simulate IBVS with for a square target comprising 4 points is placed in the world XY plane. camera pose.camera intrinsic structure (camparam) .CHAPTER 2. The camera view. can be scalar or diagonal 6 × 6 matrix (0.sph Implement spherical IBVS for point features results = sph(T) results = sph(T. image plane size and desired feature locations. Jacobian condition number.gain.sph2 Implement spherical IBVS for point features results = sph(T) results = sph(T. Two windows are shown and animated: 1. error.5) target center .2) niter eterm lambda ci depth . The params structure can be used to override simulation defaults by providing elements.depth of points to use for Jacobian. showing the target points and the camera The results structure contains time-history information about the image plane. showing the desired view (*) and the current view (o) 2. scalar for If null take actual value all points. error norm. of 4-vector.01) . The camera view.0. The camera/robot is initially at pose T and is driven to the orgin. params) Simulate IBVS with for a square target comprising 4 points is placed in the world XY plane. from simulation ([]) SEE ALSO: ibvsplot SphericalCamera.

or a vector (N × 1) for each point. See also CentralCamera. FUNCTIONS AND CLASSES 2. May 3-7 2010. error norm. of 4-vector. showing the target points and the camera The results structure contains time-history information about the image plane. IEEE Int.a stopping criteria on feature error norm (0) .01) . Conf. in Proc.visjac l.center of the target in world coords (0.gain. CentralCamera. from simulation ([]) SEE ALSO: ibvsplot SphericalCamera. I. Corke. Robotics and Automation. 5550-5555. for all points. defaults in parentheses: target size .visjac p(pt.visjac p Visual motion Jacobian for point feature J = C.the number of iterations to run the simulation (500) . scalar for If null take actual value all points. image plane size and desired feature locations. The params structure can be used to override simulation defaults by providing elements. (Anchorage). Reference “Spherical image-based visual servo and structure estimation”.CHAPTER 2. The Jacobian gives the image-plane velocity in terms of camera spatial velocity. CentralCamera. The external view. Jacobian condition number. can be scalar or diagonal 6 × 6 matrix (0. error.5) target center .0.visjac p polar. The depth of the points from the camera is given by z which is a scalar. P.the side length of the target in world units (0. z) is the image Jacobian (2N × 6) for the image plane points pt (2 × N ) described by phi (longitude) and theta (colatitude).visjac e Machine Vision Toolbox for MATLAB R 36 Copyright c Peter Corke 2011 . pp. camera pose.camera intrinsic structure (camparam) .3) niter eterm lambda ci depth .depth of points to use for Jacobian.

110. pp. Notes • SurfCornerFeature is a reference object. Computer Vision and Image Understanding (CVIU).u is a 2 × N matrix with each column the corresponding u coordinate. Tinne Tuytelaars. “SURF: Speeded Up Robust Features”. If F is a vector (N × 1) of SurfCornerFeature objects then F.CHAPTER 2. 346–359. 2008 Machine Vision Toolbox for MATLAB R 37 Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES SurfPointFeature SURF point corner feature object A subclass of PointFeature for SURF features. Luc Van Gool. Vol. No. Methods plot plot scale distance match ncc uv display char Plot feature position Plot feature scale Descriptor distance Match features Descriptor similarity Return feature coordinate Display value Convert value to string Properties u horizontal coordinate v vertical coordinate strength feature strength scale feature scale theta feature orientation [rad] descriptor feature descriptor (vector) index of image containing feature image id Properties of a vector of SurfCornerFeature objects are returned as a vector. 3. • SurfCornerFeature objects can be used in vectors and arrays Reference Herbert Bay. Andreas Ess.

FUNCTIONS AND CLASSES See also isurf. strength) as above but with speciﬁed strength.match Match SURF point features m = F. SiftPointFeature SurfPointFeature. ScalePointFeature. Options ‘thresh’. T ‘median’ match threshold (default 0.C] = F.05) Threshold at the median distance Notes • for no threshold set to [].match(f2.match(f2. options) is a vector of FeatureMatch objects that describe candidate matches between the two vectors of SURF features F and f2.CHAPTER 2. v. f = PointFeature(u. f = PointFeature(u. See also FeatureMatch Machine Vision Toolbox for MATLAB R 38 Copyright c Peter Corke 2011 .SurfPointFeature Create a SURF point feature object f = SurfPointFeature() is a point feature object with null parameters. Correspondence is based on descriptor similarity. options) as above but returns a correspodence matrix where each row contains the indices of corresponding features in F and f2 respectively. See also isurf SurfPointFeature. v) is a point feature object with speciﬁed coordinates. PointFeature. [m.

out = F.support(im. F. 0=transparent (default 0. [out.CHAPTER 2. C ‘alpha’. w) as above but if the features were extracted from an image sequence images then the feature is extracted from the appropriate image in the same sequence.support(images. 1=opaque. w) is an image of the support region of the feature F. See also SurfPointFeature Machine Vision Toolbox for MATLAB R 39 Copyright c Peter Corke 2011 .support Support region of feature out = F.plot scale Plot feature scale F. ls) as above but the optional line style arguments ls are passed to plot. w) as above but the support region is displayed.support(im. w) as above but returns the pose of the feature as a 3 × 3 homogeneous transform in SE(2) that comprises the feature position and orientation. FUNCTIONS AND CLASSES SurfPointFeature. The support region is scaled to w × w and rotated so that the feature’s orientation axis is upward. F. If F is a vector then each element is plotted. extracted from the image im in which the feature appears.plot scale(options.2) SurfPointFeature. A Indicate scale by a circle (default) Indicate scale by circle with one radial line for orientation Indicate scale and orientation by an arrow Indicate scale by a translucent disk Color of circle or disk (default green) Transparency of disk.plot scale(options) overlay a marker to indicate feature point position and scale. Options ‘circle’ ‘clock’ ‘arrow’ ‘disk’ ‘color’.T] = F.support(images.

Video AxisWebCamera.com) web camera. FUNCTIONS AND CLASSES AxisWebCamera Image from Axis webcam A concrete subclass of ImageSource that acquires images from a web camera built by Axis Communications (www. • The speciﬁed ‘resolution’ must match one that the camera is capable of. options) is an AxisWebCamera object that acquires images from an Axis Communications (www. Options ‘uint8’ ‘ﬂoat’ ‘double’ ‘grey’ ‘gamma’. S Notes: Return image with uint8 pixels (default) Return image with ﬂoat pixels Return image with double precision pixels Return greyscale image Apply gamma correction with gamma=G Subsample the image by S in both directions.CHAPTER 2.axis. Methods grab size close char Aquire and return the next image Size of image Close the image source Convert the object parameters to human readable string See also ImageSource.com). otherwise the result is not predictable.AxisWebCamera Axis web camera constructor a = AxisWebCamera(url. S ‘resolution’.axis. G ‘scale’. Machine Vision Toolbox for MATLAB R 40 Copyright c Peter Corke 2011 . Obtain an image of size S=[W H].

BagOfWords Bag of words class The BagOfWords class holds sets of features for a number of images and supports image retrieval by comparing new images with those in the ‘bag’.CHAPTER 2.display AxisWebCamera.char Convert to string A.char() is a string representing the state of the camera object in human readable form. AxisWebCamera.grab() is an image acquired from the web camera.close Close the image source A. Notes • Some web cameras have a ﬁxed picture taking interval. See also AxisWebCamera. and this function will return the most recently captured image held in the camera. FUNCTIONS AND CLASSES AxisWebCamera.grab Acquire image from the camera im = A. Machine Vision Toolbox for MATLAB R 41 Copyright c Peter Corke 2011 .close() closes the connection to the web camera.

See also PointFeature BagOfWords.org). “Video Google: a text retrieval approach to object matching in videos”. Ninth IEEE Int. 2003.Sivic and A. as produced by ISURF() for an image sequence.CHAPTER 2.1470-1477. on Computer Vision. b2) is a new bag of words created from the feature vector f but clustered to the words (and stop words) from the existing bag b2. The features are sorted into k clusters and each cluster is termed a visual word. FUNCTIONS AND CLASSES Methods isword occurrences remove stop wordvector wordfreq similarity contains exemplars display char Return all features assigned to word Return number of occurrences of word Remove stop words Return word frequency vector Return words and their frequencies Compare two word bags List the images that contain a word Display examples of word support regions Display the parameters of the bag of words Convert the parameters of the bag of words to a string Properties K nstop nimages The number of clusters speciﬁed The number of stop words speciﬁed The number of images in the bag Reference J. Notes • Uses the MEX function vl kmeans to perform clustering (vlfeat. pp. Machine Vision Toolbox for MATLAB R 42 Copyright c Peter Corke 2011 . k) is a new bag of words created from the feature vector f and with k words. Conf.Zisserman. in Proc.BagOfWords Create a BagOfWords object b = BagOfWords(f. b = BagOfWords(f. Oct. f can also be a cell array.

CHAPTER 2.char Convert to string s = B.exemplars display exemplars of words B.char() is a compact string representation of a bag of words.display Display value B. images.display() displays the parameters of the bag in a compact human readable form.char BagOfWords. The examples are displayed as a table of thumbR Machine Vision Toolbox for MATLAB 43 Copyright c Peter Corke 2011 .contains Find images containing word k = B. See also BagOfWords. BagOfWords. options) displays examples of the support regions of the words speciﬁed by the vector w.contains(w) is a vector of the indices of images in the sequence that contain one or more instances of the word w.exemplars(w. FUNCTIONS AND CLASSES See also PointFeature. Notes • This method is invoked implicitly at the command line when the result of an expression is a BagOfWords object and the command has no trailing semicolon. BagOfWords. isurf BagOfWords.

The original sequence of images from which the features were extracted must be provided as images. Machine Vision Toolbox for MATLAB R 44 Copyright c Peter Corke 2011 .occurrence(w) is the number of occurrences of the word w across all features in the bag. FUNCTIONS AND CLASSES nail images.wordfreq Word frequency statistics [w. BagOfWords. M ‘width’.isword Features from words f = B.CHAPTER 2. BagOfWords. w Number of columns to display (default 10) Maximum number of exemplars to display from any one image (default 2) Width of each thumbnail [pixels] (default 50) BagOfWords.n] = B.remove stop Remove stop words B. If w is a vector of words the result is a vector of features assigned to all the words in w.wordfreq() is a vector of word labels w and the corresponding elements of n are the number of occurrences of that word.isword(w) is a vector of feature objects that are assigned to any of the word w. All remaining words are renumbered so that the word labels are consecutive. Options ‘ncolumns’.remove stop(n) removes the n most frequent words (the stop words) from the bag. BagOfWords.occurrence Word occurrence n = B. N ‘maxperimage’.

Notes • The word vector is expensive to compute so a lazy evaluation is performed on the ﬁrst call to this function CatadioptricCamera Catadioptric camera class A concrete class for a catadioptric camera. Methods project plot hold ishold clf ﬁgure mesh point line plot camera rpy move centre delete char display project world points to image plane plot/return world point on image plane control hold for image plane test ﬁgure hold for image plane clear image plane ﬁgure holding the image plane draw shape represented as a mesh draw homogeneous points on image plane draw homogeneous lines on image plane draw camera set camera attitude copy of Camera after motion get world coordinate of camera centre object destructor convert camera parameters to string display camera parameters Machine Vision Toolbox for MATLAB R 45 Copyright c Peter Corke 2011 .CHAPTER 2.wordvector Word frequency vector wf = B.wordvector(J) is the word frequency vector for the J’th image in the bag. The vector is K × 1 and the angle between any two WFVs is an indication of image similarity. FUNCTIONS AND CLASSES BagOfWords. subclass of Camera.

C = CatadioptricCamera(options) as above but with speciﬁed parameters.CHAPTER 2.CatadioptricCamera Create central projection camera object C = CatadioptricCamera() creates a central projection camera with canonic parameters: f=1 and name=’canonic’. FUNCTIONS AND CLASSES Properties (read/write) npix pp rho f p T image dimensions in pixels (2 × 1) intrinsic: principal point (2 × 1) intrinsic: pixel dimensions (2 × 1) [metres] intrinsic: focal length [metres] intrinsic: tangential distortion parameters extrinsic: camera pose as homogeneous transformation Properties (read only) nu nv u0 v0 number of pixels in u-direction number of pixels in v-direction principal point u-coordinate principal point v-coordinate Notes • Camera is a reference object. • Camera objects can be used in vectors and arrays See also CentralCamera. Camera CatadioptricCamera. Machine Vision Toolbox for MATLAB R 46 Copyright c Peter Corke 2011 .

CHAPTER 2. K ‘maxangle’. SphericalCamera CatadioptricCamera.and v-axes parallel to x. options) are the image plane coordinates for the world points p. ‘stereographic’ Parameter for the projection model The maximum viewing angle above the horizontal plane. Set the camera pose to the homogeneous transformation T before projecting points to the camera image plane. S ‘noise’. 10um pixels. A ‘resolution’. T ‘Tcam’. F ‘default’ ‘projection’. Temporarily overrides the current camera pose C.plot Machine Vision Toolbox for MATLAB R 47 Copyright c Peter Corke 2011 . ‘sine’. Catadioptric model: ‘equiangular’ (default). M ‘k’. f=8mm. Image plane resolution: N × N or N=[W H]. P ‘pixel’. N ‘sensor’.and y-axes respectively.project(p. u. CatadioptricCamera. SIGMA ‘pose’. camera at origin.T. See also Camera. Image sensor size in metres (2 × 1) Principal point (2 × 1) Pixel size: S × S or S=[W H]. T Name of camera Focal length (metres) Default camera parameters: 1024 × 1024. ﬁsheyecamera. N ‘focal’. ‘equisolid’. T Transform all points by the homogeneous transformation T before projecting them to the camera image plane. Standard deviation of additive Gaussian noise added to returned image projections Pose of the camera as a homogeneous transformation Notes • The elevation angle range is from -pi/2 (below the mirror) to maxangle above the horizontal plane. The columns of p (3 × N ) are the world points and the columns of uv (2 × N ) are the corresponding image plane points. FUNCTIONS AND CLASSES Options ‘name’.project Project world points to image plane uv = C. S ‘centre’. See also Camera. Options ‘Tobj’. optical axis is z-axis.

FUNCTIONS AND CLASSES FeatureMatch Feature correspondence object This class represents the correspondence between two PointFeature objects. SiftPointFeature Machine Vision Toolbox for MATLAB R 48 Copyright c Peter Corke 2011 . If F is a vector (N × 1) of FeatureMatch objects then F. • FeatureMatch objects can be used in vectors and arrays • Operates with all objects derived from PointFeature. SurfPointFeature and SiftPointFeature. SurfPointFeature.CHAPTER 2. Note • FeatureMatch is a reference object. such as ScalePointFeature. See also PointFeature. Methods plot show ransac inlier outlier subset display char Plot corresponding points Show summary statistics of corresponding points Determine inliers and outliers Return inlier matches Return outlier matches Return a subset of matches Display value of match Convert value of match to string Properties p1 Point coordinates in view 1 (2 × 1) p2 Point coordinates in view 2 (2 × 1) p Point coordinates in view 1 and 2 (4 × 1) distance Match strength between the points Properties of a vector of FeatureMatch objects are returned as a vector. A vector of FeatureMatch objects can represent the correspondence between sets of points.p1 is a 2 × N matrix with each column the corresponding view 1 point coordinate.

one per element.char Convert to string s = M. m = FeatureMatch(f1. Notes • This method is invoked implicitly at the command line when the result of an expression is a FeatureMatch object and the command has no trailing semicolon. If M is a vector then the elements are printed one per line. SiftPointFeature FeatureMatch. See also PointFeature.display Display value M.CHAPTER 2. See also FeatureMatch. f2) as above but the strength is set to NaN.char() is a compact string representation of the match object. FeatureMatch. If M is a vector then the string has multiple lines. f2. SurfPointFeature.display() displays a compact human-readable representation of the feature pair.FeatureMatch Create a new FeatureMatch object m = FeatureMatch(f1. s) is a new FeatureMatch object describing a correspondence between point features f1 and f2 with a strength of s. Notes • Only the coordinates of the PointFeature are kept. FUNCTIONS AND CLASSES FeatureMatch.char Machine Vision Toolbox for MATLAB R 49 Copyright c Peter Corke 2011 .

outlier. Notes • Outliers are not determined until after RANSAC is run.inlier Inlier features m2 = M.CHAPTER 2. Notes • Inliers are not determined until after RANSAC is run.ransac FeatureMatch.p Feature point coordinate pairs p = M. FeatureMatch. See also FeatureMatch. FUNCTIONS AND CLASSES FeatureMatch.v2].outlier() is a subset of the FeatureMatch vector M that are considered to be outliers. See also FeatureMatch.inlier. FeatureMatch.v1.outlier Outlier features m2 = M.u2. Each column contains the coordinates of a pair of corresponding points [u1.inlier() is a subset of the FeatureMatch vector M that are considered to be inliers.p() is a 4 × N matrix containing the feature point coordinates. Machine Vision Toolbox for MATLAB R 50 Copyright c Peter Corke 2011 .ransac FeatureMatch.

The ﬁgure must comprise views 1 and 2 side by side. FeatureMatch. FUNCTIONS AND CLASSES See also FeatureMatch.plot() M.p1.plot Show corresponding points M. FeatureMatch. for example by: idisp({im1. These are the (u.FeatureMatch.p1() is a 2 × N matrix containing the feature points coordinates from view 1.p FeatureMatch. FeatureMatch. These are the (u.p1 Feature point coordinates from view 1 p = M.plot() overlays the correspondences in the FeatureMatch vector M on the current ﬁgure.p2() is a 2 × N matrix containing the feature points coordinates from view 1.FeatureMatch.p2 FeatureMatch.p1.plot(ls) as above but the optional line style arguments ls are passed to plot. See also FeatureMatch. Machine Vision Toolbox for MATLAB R 51 Copyright c Peter Corke 2011 .CHAPTER 2. FeatureMatch.v) properties of the feature F2 passed to the constructor.p2 Feature point coordinates from view 2 p = M. See also FeatureMatch.p2. FeatureMatch.v) properties of the feature F1 passed to the constructor.im2}) m.p FeatureMatch.

ransac(func. 1e-4). See also idisp FeatureMatch. m.ransac Apply RANSAC M. inliers and outliers (and their percentages). f2 = isurf(im2). Machine Vision Toolbox for MATLAB R 52 Copyright c Peter Corke 2011 . ransac FeatureMatch. m = f1. The options are passed to the RANSAC() function.match(f2). Elements of the FeatureMatch vector have their status updated in place to indicate whether they are inliers or outliers. See also fmatrix. options) applies the RANSAC algorithm to ﬁt the point correspondences to the model described by the function func. homography. and an error is created if this UserData is not found.show() is a compact summary of the FeatureMatch vector M that gives the number of matches. Example f1 = isurf(im1). FUNCTIONS AND CLASSES Notes • Using IDISP as above adds UserData to the ﬁgure.ransac( @fmatrix.CHAPTER 2.show Display summary statistics of the FeatureMatch vector M.

CHAPTER 2. X | | | + (principal point) | | Z-axis is into the page. Y This camera model assumes central projection.subset Subset of matches m2 = M. Methods project plot hold ishold clf ﬁgure mesh point line plot camera rpy move centre delete char display project world points to image plane plot/return world point on image plane control hold for image plane test ﬁgure hold for image plane clear image plane ﬁgure holding the image plane draw shape represented as a mesh draw homogeneous points on image plane draw homogeneous lines on image plane draw camera set camera attitude copy of Camera after motion get world coordinate of camera centre object destructor convert camera parameters to string display camera parameters Machine Vision Toolbox for MATLAB R 53 Copyright c Peter Corke 2011 . the focal point is at z=0 and the image plane is at z=f.subset(n) is a FeatureMatch vector with no more than n elements sampled uniformly from M. v. The camera coordinate system is: 0------------> u. that is. FishEyeCamera Fish eye camera class A concrete class a ﬁsheye lense projection camera. The image is not inverted. FUNCTIONS AND CLASSES FeatureMatch.

SIGMA ‘pose’. N ‘default’ ‘projection’.FishEyeCamera Create ﬁsheyecamera object C = FishEyeCamera() creates a ﬁsheye camera with canonic parameters: f=1 and name=’canonic’. ‘equisolid’. Options ‘name’. T Name of camera Default camera parameters: 1024 × 1024. ‘stereographic’ Parameter for the projection model Image plane resolution: N × N or N=[W H]. camera at origin.and y. u. C = FishEyeCamera(options) as above but with speciﬁed parameters. FUNCTIONS AND CLASSES Properties (read/write) npix pp f rho T image dimensions in pixels (2 × 1) intrinsic: principal point (2 × 1) intrinsic: focal length [metres] intrinsic: pixel dimensions (2 × 1) [metres] extrinsic: camera pose as homogeneous transformation Properties (read only) nu nv number of pixels in u-direction number of pixels in v-direction Notes • Camera is a reference object. Fisheye model: ‘equiangular’ (default). Image sensor size [metres] (2 × 1) Principal point (2 × 1) Pixel size: S × S or S=[W H]. M ‘k’. P ‘pixel’. S ‘noise’. K ‘resolution’. optical axis is z-axis. ‘sine’.CHAPTER 2. f=8mm. Standard deviation of additive Gaussian noise added to returned image projections Pose of the camera as a homogeneous transformation R Machine Vision Toolbox for MATLAB 54 Copyright c Peter Corke 2011 .and v-axes are parallel to x. • Camera objects can be used in vectors and arrays See also Camera FishEyeCamera. N ‘sensor’.axes respectively. 10um pixels. S ‘centre’.

For every edge pixel in the input image a set of cells in the Hough accumulator (voting array) are incremented. options) are the image plane coordinates for the world points p. CentralCamera.plot Hough Hough transform class The Hough transform is a technique for ﬁnding lines in an image using a voting scheme. The columns of p (3 × N ) are the world points and the columns of uv (2 × N ) are the corresponding image plane points.project(p.CHAPTER 2. In this version of the Hough transform lines are described by: d = y cos(theta) + x sin(theta) Machine Vision Toolbox for MATLAB R 55 Copyright c Peter Corke 2011 . See also FishEyeCamera. SphericalCamera FishEyeCamera. T ‘Tcam’.T. CatadioptricCamera. Options ‘Tobj’. Set the camera pose to the homogeneous transformation T before projecting points to the camera image plane. T Transform all points by the homogeneous transformation T before projecting them to the camera image plane.project Project world points to image plane uv = C. Temporarily overrides the current camera pose C. FUNCTIONS AND CLASSES Notes • If K is not speciﬁed it is computed such that the circular imaging region maximally ﬁlls the square image plane. See also Camera.

Hough Create Hough transform object ht = Hough(E. with columns corresponding to theta and rows corresponding to offset (d). The voting array is 2-dimensional. Methods plot show lines char display Overlay detected lines Display the Hough accumulator Return line features Convert Hough parameters to string Display Hough parameters Properties Nrho Ntheta A rho theta edgeThresh houghThresh suppress interpWidth Number of bins in rho direction Number of bins in theta direction The Hough accumulator (Nrho x Ntheta) rho values for the centre of each bin vertically Theta values for the centre of each bin horizontally Threshold on relative edge pixel strength Threshold on relative peak strength Radius of accumulator cells cleared around peak Width of region used for peak interpolation Notes • Hough is a reference object. a vertical line has theta = pi/2 or -pi/2.CHAPTER 2. A horizontal line has theta = 0. Theta spans the range -pi/2 to pi/2 in Ntheta steps. By default the vote is incremented by Machine Vision Toolbox for MATLAB R 56 Copyright c Peter Corke 2011 . options) is the Hough transform of the edge image E. For every pixel in the edge image E (H ×W ) greater than a threshold the corresponding elements of the accumulator are incremented. FUNCTIONS AND CLASSES where theta is the angle the line makes to horizontal axis. and d is the perpendicular distance between (0. See also LineFeature Hough.H). Offset is in the range -rho max to rho max where rho max=max(W.0) and the line.

edgeThresh. Hough. Default 400 × 401. Set ht. W ‘houghthresh’.char() is a compact string representation of the Hough transform parameters. Nrho].char Convert to string s = HT. See also Hough. W ‘nbins’.5) Set ht.edgeThresh (default 0. else N = [Ntheta.suppress (default 0) Set number of bins. The threshold is determined from the maximum edge strength value x ht. T ‘suppress’.display() displays a compact human-readable string representation of the Hough transform parameters. if N is scalar set Nrho=Ntheta=N. otherwise the edge pixel value is the vote strength Interpolation width (default 3) Set ht.houghThresh (default 0. Notes • This method is invoked implicitly at the command line when the result of an expression is a Hough object and the command has no trailing semicolon. N All edge pixels have equal weight. Options ‘equal’ ‘interpwidth’.CHAPTER 2. FUNCTIONS AND CLASSES the edge strength but votes can be made equal with the option ‘equal’. T ‘edgethresh’.char Machine Vision Toolbox for MATLAB R 57 Copyright c Peter Corke 2011 . Hough.1).display Display value HT.

The peak detection loop breaks early if the remaining peak has a strength less than HT. LineFeature Hough.plot Plot line features HT. Lines are the coordinates of peaks in the Hough accumulator.show Display the Hough accumulator as image s = HT.plot() as above but returns a vector of graphics handles for each line. See also Hough. reﬁned to subpixel precision.lines Hough.plot. Machine Vision Toolbox for MATLAB R 58 Copyright c Peter Corke 2011 .plot(n. H = HT.plot() overlays all detected lines on the current ﬁgure. then all elements in an HT.lines() is a vector of LineFeature objects that represent the dominant lines in the Hough accumulator. L = HT.show() displays the Hough vote accumulator as an image using the hot colormap.plot(n) overlays a maximum of n strongest lines on the current ﬁgure. ls) as above but the optional line style arguments ls are passed to plot.lines Find lines L = HT. The highest peak is found. where ‘heat’ is proportional to the number of votes. See also Hough. HT.lines(n) as above but returns no more than n LineFeature objects.houghThresh times the maximum vote value. HT. FUNCTIONS AND CLASSES Hough.CHAPTER 2.suppress radius around are zeroed so as to eliminate multiple close minima. The process is repeated for all peaks.

CHAPTER 2. Methods plot seglength display char Plot the line segment Determine length of line segment Display value Convert value to string Properties rho Offset of the line theta Orientation of the line strength Feature strength length Length of the line Properties of a vector of LineFeature objects are returned as a vector. hot LineFeature Line feature class This class represents a line feature.rho is an N × 1 vector of the rho element of each feature. • LineFeature objects can be used in vectors and arrays See also Hough. Note • LineFeature is a reference object. FUNCTIONS AND CLASSES See also colormap. RegionFeature. PointFeature Machine Vision Toolbox for MATLAB R 59 Copyright c Peter Corke 2011 . If L is a vector (N × 1) of LineFeature objects then L.

L = LineFeature(rho. If L is a vector then the elements are printed one per line.char Convert to string s = L. strength. LENGTH is undeﬁned.LineFeature Create a line feature object L = LineFeature() is a line feature object with null parameters. FUNCTIONS AND CLASSES LineFeature. Notes • This method is invoked implicitly at the command line when the result of an expression is a LineFeature object and the command has no trailing semicolon. theta.display Display value L. LineFeature.char() is a compact string representation of the line feature. one per element.char Machine Vision Toolbox for MATLAB R 60 Copyright c Peter Corke 2011 .CHAPTER 2. L = LineFeature(l2) is a deep copy of the line feature l2. LineFeature. strength) is a line feature object with the speciﬁed properties. See also LineFeature.display() displays a compact human-readable representation of the feature. L = LineFeature(rho. length) is a line feature object with the speciﬁed properties. theta. If L is a vector then the string has multiple lines.

See also icanny Machine Vision Toolbox for MATLAB R 61 Copyright c Peter Corke 2011 .seglength Compute length of line segments The Hough transform identiﬁes lines but cannot determine their length. gap) is a copy of the line feature object with the property length updated to the length of the line (pixels).plot Plot line L. L. less than gap pixels are tolerated.CHAPTER 2. This method examines the edge pixels in the original image and determines the longest stretch of non-zero pixels along the line. l2 = L. Small gaps. FUNCTIONS AND CLASSES LineFeature.plot() overlay the line on current plot. See also icanny LineFeature.points Return points on line segments p = L.seglength(edge) as above but the maximum allowable gap is 5 pixels.points(edge) is the set of points that lie along the line in the edge image edge are determined. Notes • If L is a vector then each element is plotted.plot(ls) as above but the optional line style arguments ls are passed to plot. LineFeature. l2 = L.seglength(edge.

char() is a string representing the state of the movie object in human readable form.close Close the image source M. options) is an Movie object that returns frames from the movie ﬁle ﬁle. Machine Vision Toolbox for MATLAB R 62 Copyright c Peter Corke 2011 .com). FUNCTIONS AND CLASSES Movie Class to read movie ﬁle A concrete subclass of ImageSource that acquires images from a web camera built by Axis Communications (www.CHAPTER 2. Options ‘uint8’ ‘ﬂoat’ ‘double’ ‘grey’ ‘gamma’. Movie. G ‘scale’.close() closes the connection to the movie. S ‘skip’. S Return image with uint8 pixels (default) Return image with ﬂoat pixels Return image with double precision pixels Return greyscale image Apply gamma correction with gamma=G Subsample the image by S in both directions Read every S’th frame from the movie Movie.char Convert to string M.Movie Image source constructor m = Movie(ﬁle. Movie.axis.

Machine Vision Toolbox for MATLAB R 63 Copyright c Peter Corke 2011 . F Skip frames. planar. undirected graph create an n-d. and return current+S frame Return frame F within the movie Notes • If no output argument given the image is displayed using IDISP. FUNCTIONS AND CLASSES Movie. vid • edges are represented by integer ids. PGraph Simple graph class g = PGraph() g = PGraph(n) create a 2D. eid Graph connectivity is maintained by a labeling algorithm and this is updated every time an edge is added.grab(options) as above but allows the next frame to be speciﬁed. undirected graph Graphs • are undirected • are symmetric cost edges (A to B is same cost as B to A) • are embedded in coordinate system • have no loops (edges from A to A) • vertices are represented by integer ids.grab() acquires the next image from the movie im = M. S ‘frame’.CHAPTER 2. Options ‘skip’.grab Acquire next frame from movie im = M.

path(v) set goal vertex. return vid add edge from v1 to v2.CHAPTER 2.distances(coord) return sorted distances from coord and vertices To change the distance metric create a subclass of PGraph and override the method distance metric().edges(e) g.clear() add vertex.add node(coord) g. return vid add vertex and edge to v.cost(e) g.component(v) g.add edge(v1. FUNCTIONS AND CLASSES Methods Constructing the graph g.next(v) g.closest(coord) return vertex closest to coord g. v) g. v2) g. Object properties (read/write) g.n number of nodes Machine Vision Toolbox for MATLAB R 64 Copyright c Peter Corke 2011 . return eid remove all nodes and edges from the graph Information from graph g.coord(v) g.connectivity() g.plot() g.distance(v1. and plan paths return d of neighbour of v closest to goal return list of nodes from v to goal Graph and world points g.pick() char(g) return vid for edge return cost for edge list return coordinate of node v return vid for edge return component id for vertex return number of edges for all nodes set goal vertex for path planning return vertex id closest to picked point display summary info about the graph Planning paths through the graph g. v2) distance between v1 and v2 as the crow ﬂies g.neighbours(v) g.goal(v) g.add node(coord.

v) adds a node with coordinate x and connected to node v by an edge. Machine Vision Toolbox for MATLAB R 65 Copyright c Peter Corke 2011 . v = G.add node(x) adds a node with coordinate x. v. FUNCTIONS AND CLASSES PGraph. PGraph. C) adds a node with coordinate x and connected to node v by an edge with cost C. options) returns a graph object embedded in d dimensions.CHAPTER 2. PGraph. E = G. v = G.add node(x. and returns the edge id E. and returns the node id v.add node(x.PGraph Graph class constructor g = PGraph(d.add edge Add an edge to the graph E = G. C) add an edge between nodes with id v1 and v2 with cost C.add edge(v1. v2) add an edge between nodes with id v1 and v2. M ‘verbose’ Use the distance metric M for path planning Specify verbose operation Note • The distance metric is either ‘Euclidean’ or ‘SE2’ which is the sum of the squares of the difference in position and angle modulo 2pi. where x is D × 1.add node Add a node to the graph v = G. Options ‘distance’.add edge(v1. v2.

CHAPTER 2. PGraph.coord(v) return coordinate vector.CLEAR() removes all nodes and edges.CLOSEST(x) return id of node geometrically closest to coordinate x. edges and components.closest(x) return id of node geometrically closest to coordinate x.char() returns a compact human readable representation of the state of the graph including the number of vertices.coord Coordinate of node x = G. PGraph. Machine Vision Toolbox for MATLAB R 66 Copyright c Peter Corke 2011 .closest Find closest node v = G. of node id v. D × 1.d] = G. FUNCTIONS AND CLASSES PGraph. [v.char Convert graph to string s = G.connectivity Graph connectivity C = G.connectivity() returns the total number of edges in the graph. PGraph.clear Clear the graph G. and the distance d. PGraph.

CHAPTER 2. FUNCTIONS AND CLASSES

PGraph.cost

Cost of edge

C = G.cost(E) return cost of edge id E.

PGraph.display

Display state of the graph

G.display() displays a compact human readable representation of the state of the graph including the number of vertices, edges and components.

See also

PGraph.char

PGraph.distance

Distance between nodes

d = G.distance(v1, v2) return the geometric distance between the nodes with id v1 and v2.

PGraph.distances

distance to all nodes

d = G.distances(v) returns vector of geometric distance from node id v to every other node (including v) sorted into increasing order by d. [d,w] = G.distances(v) returns vector of geometric distance from node id v to every other node (including v) sorted into increasing order by d where elements of w are the corresponding node id.

Machine Vision Toolbox for MATLAB

R

67

Copyright c Peter Corke 2011

CHAPTER 2. FUNCTIONS AND CLASSES

PGraph.edges

Find edges given vertex

E = G.edges(v) return the id of all edges from node id v.

PGraph.goal

Set goal node

G.goal(vg) for least-cost path through graph set the goal node. The cost of reaching every node in the graph connected to vg is computed.

See also

PGraph.path cost is total distance from goal

PGraph.neighbours

Neighbours of a node

n = G.neighbours(v) return a vector of ids for all nodes which are directly connected neighbours of node id v. [n,C] = G.neighbours(v) return a vector n of ids for all nodes which are directly connected neighbours of node id v. The elements of C are the edge costs of the paths to the corresponding node ids in n.

PGraph.next

Find next node toward goal

v = G.next(vs) return the id of a node connected to node id vs that is closer to the goal.

See also

PGraph.goal, PGraph.path

Machine Vision Toolbox for MATLAB

R

68

Copyright c Peter Corke 2011

CHAPTER 2. FUNCTIONS AND CLASSES

PGraph.path

Find path to goal node

p = G.path(vs) return a vector of node ids that form a path from the starting node vs to the previously speciﬁed goal. The path includes the start and goal node id.

See also

PGraph.goal

PGraph.pick

Graphically select a node

v = G.pick() returns the id of the node closest to the point clicked by user on a plot of the graph.

See also

PGraph.plot

PGraph.plot

Plot the graph

G.plot(opt) plot the graph in the current ﬁgure. Nodes are shown as colored circles.

Options

‘labels’ ‘edges’ ‘edgelabels’ ‘MarkerSize’, S ‘MarkerFaceColor’, C ‘MarkerEdgeColor’, C ‘componentcolor’ Display node id (default false) Display edges (default true) Display edge id (default false) Size of node circle Node circle color Node circle edge color Node color is a function of graph component

Machine Vision Toolbox for MATLAB

R

69

Copyright c Peter Corke 2011

CHAPTER 2. FUNCTIONS AND CLASSES

PGraph.showComponent

t

G.showcomponent(C) plots the nodes that belong to graph component C.

PGraph.showVertex

Highlight a vertex

G.showVertex(v) highlights the vertex v with a yellow marker.

PGraph.vertices

Find vertices given edge

v = G.vertices(E) return the id of the nodes that deﬁne edge E.

PointFeature

PointCorner feature object

A superclass for image corner features.

Methods

plot distance ncc uv display char Plot feature position Descriptor distance Descriptor similarity Return feature coordinate Display value Convert value to string

Machine Vision Toolbox for MATLAB

R

70

Copyright c Peter Corke 2011

display() displays a compact human-readable representation of the feature. SiftPointFeature PointFeature. PointFeature. If F is a vector then the elements are printed one per line.char() is a compact string representation of the point feature. f = PointFeature(u. strength) as above but with speciﬁed strength.PointFeature Create a point feature object f = PointFeature() is a point feature object with null parameters. If F is a vector (N × 1) of PointFeature objects then F. v) is a point feature object with speciﬁed coordinates. PointFeature. SurfPointFeature. If F is a vector then the string has multiple lines. Notes • This method is invoked implicitly at the command line when the result of an expression is a PointFeature object and the command has no trailing semicolon.u is a 2 × N matrix with each column the corresponding point coordinate. f = PointFeature(u.display Display value F.CHAPTER 2. one per element. Machine Vision Toolbox for MATLAB R 71 Copyright c Peter Corke 2011 .char Convert to string s = F. FUNCTIONS AND CLASSES Properties u horizontal coordinate v vertical coordinate strength feature strength descriptor feature descriptor (vector) Properties of a vector of PointFeature objects are returned as a vector. See also ScalePointFeature. v.

distance Distance between feature descriptors d = F. options) is a vector of FeatureMatch objects that describe candidate matches between the two vectors of point features F and f2.05) Threshold at the median distance See also FeatureMatch PointFeature. FUNCTIONS AND CLASSES See also PointFeature. [m. If F is a vector then D is a vector whose elements are the distance between the corresponding element of F and f1.char PointFeature. If F is a vector then d is a vector whose elements are the distance between the corresponding element of F and f1.match(f2. where 1 is perfect match.match(f2.CHAPTER 2. options) as above but returns a correspodence matrix where each row contains the indices of corresponding features in F and f2 respectively. the norm of the Euclidean distance.distance(f1) is the distance between feature descriptors. Options ‘thresh’. Machine Vision Toolbox for MATLAB R 72 Copyright c Peter Corke 2011 .C] = F. PointFeature.ncc(f1) is the similarty between feature descriptors which is a scalar in the interval -1 to 1. T ‘median’ match threshold (default 0.ncc Feature descriptor similarity s = F.match Match point features m = F.

Methods plot area moments centroid perimeter transform inside intersection difference union xor display char Plot polygon Area of polygon Moments of polygon Centroid of polygon Perimter of polygon Transform polygon Test if points are inside polygon Intersection of two polygons Difference of two polygons Union of two polygons Exclusive or of two polygons print the polygon in human readable form convert the polgyon to human readable string Properties vertices extent n List of polygon vertices. FUNCTIONS AND CLASSES PointFeature.plot(ls) as above but the optional line style arguments ls are passed to plot.General polygon class p = Polygon(vertices).plot Plot feature F.CHAPTER 2. If F is a vector then each element is plotted.plot() overlay a marker at the feature position. Polygon . one per column Bounding box [minx maxx. miny maxy] Number of vertices Notes • this is reference class object • Polygon objects can be used in vectors and arrays R Machine Vision Toolbox for MATLAB 73 Copyright c Peter Corke 2011 . F.

HEIGHT].centroid Centroid of polygon x = P.char() is a compact representation of the polgyon in human readable form. intersection. so use with care.mit. Machine Vision Toolbox for MATLAB R 74 Copyright c Peter Corke 2011 . kirill@plume. p = Polygon(C. wh) is a rectangle centred at C with dimensions wh=[WIDTH. difference.area() is the area of the polygon.mit. http://puddle.centroid() is the centroid of the polygon.edu/ glenn/kirill/saga. Polygon.char String representation s = P.html and require a licence. FUNCTIONS AND CLASSES Acknowledgement The methods inside. one column per vertex.Polygon Polygon class constructor p = Polygon(v) is a polygon with vertices given by v. Polygon. Polygon.area Area of polygon a = P. Pankratov.edu.CHAPTER 2. However the author does not respond to email regarding the licence. union. Polygon. and xor are based on code written by: Kirill K.

display() displays the polygon in a compact human readable form. returns coordinates of P.inside(p) tests if points given by columns of p are inside the polygon. FUNCTIONS AND CLASSES Polygon. Notes • If polygons P and q are not intersecting.difference(q) is polygon P minus polygon q. The corresponding elements of in are either true or false. • If the result d is not simply connected or consists of several polygons.difference Difference of polygons d = P.CHAPTER 2.intersect(plist) indicates whether or not the Polygon P intersects with i(j) = 1 if p intersects polylist(j).inside Test if points are inside polygon in = p. See also Polygon. Machine Vision Toolbox for MATLAB R 75 Copyright c Peter Corke 2011 .display Display polygon P. resulting vertex list will contain NaNs.intersect Intersection of polygon with list of polygons i = P. Polygon. else 0. Polygon.char Polygon.

Polygon. Notes • If these polygons are not intersecting. i is an N × 2 matrix with one column per intersection.moments(p.perimeter Perimeter of polygon L = P.moments Moments of polygon a = P. y1 y2].perimeter() is the perimeter of the polygon.intersection(q) is a Polygon representing the intersection of polygons P and q.CHAPTER 2.intersect line Intersection of polygon and line segment i = P. q) is the pq’th moment of the polygon. FUNCTIONS AND CLASSES Polygon. each column is [x y]’.intersection Intersection of polygons i = P. • If intersection consist of several disjoint polygons (for non-convex P or q) then vertices of i is the concatenation of the vertices of these polygons. Polygon.intersect line(L) is the intersection points of a polygon P with the line segment L=[x1 x2. Machine Vision Toolbox for MATLAB R 76 Copyright c Peter Corke 2011 . returns empty polygon. See also mpq poly Polygon.

Polygon. returns a polygon with vertices of both polygons separated by NaNs. P. FUNCTIONS AND CLASSES Polygon.transform Transformation of polygon vertices p2 = P.union(q) is a Polygon representing the union of polygons P and q.transform(T) is a new Polygon object whose vertices have been transfored by the 3 × 3 homgoeneous transformation T.plot(ls) as above but pass the arguments ls to plot.CHAPTER 2.xor Exclusive or of polygons i = P. returns a polygon with vertices of both polygons separated by NaNs. Machine Vision Toolbox for MATLAB R 77 Copyright c Peter Corke 2011 . • If the result P is not simply connected (such as a polygon with a “hole”) the resulting contour consist of counter.plot Plot polygon P. Notes • If these polygons are not intersecting. Polygon. Polygon. Notes • If these polygons are not intersecting.plot() plot the polygon.union(q) is a Polygon representing the union of polygons P and q.union Union of polygons i = P.clockwise “outer boundary” and one or more clock-wise “inner boundaries” around “holes”.

clockwise “outer boundary” and one or more clock-wise “inner boundaries” around “holes”. d) is a new Ray3D object deﬁned by a point on the ray p0 and a direction vector d. FUNCTIONS AND CLASSES • If the result P is not simply connected (such as a polygon with a “hole”) the resulting contour consist of counter. Methods intersect closest char display Intersection of ray with plane or ray Closest distance between point and ray Ray parameters as human readable string Display ray parameters in human readable form Properties P0 d A point on the ray (3 × 1) Direction of the ray. deﬁned by a point on the ray and a direction unit-vector. Ray3D Ray in 3D space This object represents a ray in 3D space.CHAPTER 2. Machine Vision Toolbox for MATLAB R 78 Copyright c Peter Corke 2011 . unit vector (3 × 1) Notes • Ray3D objects can be used in vectors and arrays Ray3D.Ray3D Ray constructor R = Ray3D(p0.

char Convert to string s = R. See also Ray3D.char() is a compact string representation of the Ray3D’s value.CHAPTER 2.closest(p) is the point on the ray R closest to the point p.closest(p) as above but also returns the distance E between x and p. Ray3D. Ray3D.display Display value R. corresponding to the intersection of R(i) with r2. Machine Vision Toolbox for MATLAB R 79 Copyright c Peter Corke 2011 . If R is a vector then the elements are printed one per line.closest Closest distance between point and ray x = R.intersect(r2) is the point on R that is closest to the ray r2. [x. If R is a vector then then x has multiple columns.char Ray3D.intersect(r2) as above but also returns the closest distance between the rays.E] = R. FUNCTIONS AND CLASSES Ray3D.E] = R.display() displays a compact human-readable representation of the Ray3D’s value. [x.intersect Intersetion of ray with line or plane x = R. Notes • This method is invoked implicitly at the command line when the result of an expression is a Ray3D object and the command has no trailing semicolon. one per element. If R is a vector then the string has multiple lines.

maximum vertical coordinate the number of pixels the value of the pixels forming this region the label assigned to this region a list of indices of features that are children of this feature coordinate of a point on the perimeter a list of edge points 2 × N matrix number of edge pixels true if region touches edge of the image major axis length of equivalent ellipse minor axis length of equivalent ellipse angle of major ellipse axis to horizontal axis aspect ratio b/a (always <= 1. Methods boundary box plot plot boundary plot box plot ellipse display char Return the boundary as a list Return the bounding box Plot the centroid Plot the boundary Plot the bounding box Plot the equivalent ellipse Display value Convert value to string Properties uc vc umin umax vmin vmax area class label children edgepoint edge perimeter touch a b theta shape circularity moments centroid. less for other shapes a structure containing moments of order 0 to 2 Machine Vision Toolbox for MATLAB R 80 Copyright c Peter Corke 2011 .b. corresponding to the intersection of R(i) with p. maximum horizontal coordinate bounding box.CHAPTER 2. horizontal coordinate centroid. minimum horizontal coordinate bounding box. minimum vertical coordinate bounding box. If R is a vector then x has multiple columns.intersect(p) returns the point of intersection between the ray R and the plane p=(a.d) where aX + bY + cZ + d = 0. RegionFeature Region feature class This class represents a region feature.0) 1 for a circle. vertical coordinate bounding box. FUNCTIONS AND CLASSES x = R.c.

RegionFeature. one per element. See also iblobs. ymin. • RegionFeature objects can be used in vectors and arrays • This class behaves differently to LineFeature and PointFeature when getting properties of a vector of RegionFeature objects. For example R.char() is a compact string representation of the region feature. ymax].boundary Boundary in polar form [d.box() is the bounding box in standard Toolbox form [xmin. imoments RegionFeature.CHAPTER 2.th] = R.RegionFeature Create a region feature object R = RegionFeature() is a region feature object with null parameters. d(i) and th(i) are the distance to the boundary point and the angle respectively. RegionFeature.boundary() is a polar representation of the boundary with respect to the centroid. FUNCTIONS AND CLASSES Note • RegionFeature is a reference object.uc will be a list not a vector. These vectors have 400 elements irrespective of region size.xmax. RegionFeature.char Convert to string s = R. Machine Vision Toolbox for MATLAB R 81 Copyright c Peter Corke 2011 .box Return bounding box b = R. If R is a vector then the string has multiple lines.

plot boundary(ls) as above but the optional line style arguments ls are passed to plot. If R is a vector then the elements are printed one per line.plot(ls) as above but the optional line style arguments ls are passed to plot. Notes • this method is invoked implicitly at the command line when the result of an expression is a RegionFeature object and the command has no trailing semicolon. R. Notes • If R is a vector then each element is plotted. It is indicated with overlaid o. If R is a vector then each element is plotted.plot boundary() overlay perimeter points on current plot.plot boundary plot boundary R.char RegionFeature.plot Plot centroid R.display() is a compact string representation of the region feature.and xmarkers. See also RegionFeature. R. RegionFeature. FUNCTIONS AND CLASSES RegionFeature.CHAPTER 2. Machine Vision Toolbox for MATLAB R 82 Copyright c Peter Corke 2011 .display Display value R.plot() overlay the centroid on current plot.

FUNCTIONS AND CLASSES See also boundmatch RegionFeature. Methods plot plot scale distance ncc uv display char Plot feature position Plot feature scale Descriptor distance Descriptor similarity Return feature coordinate Display value Convert value to string Machine Vision Toolbox for MATLAB R 83 Copyright c Peter Corke 2011 .plot box() overlay the the bounding box of the region on current plot. ScalePointFeature ScalePointCorner feature object A subclass of PointFeature for features with scale.CHAPTER 2. If R is a vector then each element is plotted.plot ellipse(ls) as above but the optional line style arguments ls are passed to plot. R.plot ellipse Plot equivalent ellipse R. If R is a vector then each element is plotted. RegionFeature. R.plot ellipse() overlay the the equivalent ellipse of the region on current plot.plot box(ls) as above but the optional line style arguments ls are passed to plot.plot box Plot bounding box R.

strength) as above but with speciﬁed strength.CHAPTER 2. ScalePointFeature. If F is a vector then each element is plotted. 0=transparent (default 0.plot scale(options) overlay a marker at the feature position. See also PointFeature. C ‘alpha’.ScalePointFeature Create a scale point feature object f = ScalePointFeature() is a point feature object with null parameters. ls) as above but the optional line style arguments ls are passed to plot. SurfPointFeature. 1=opaque. f = ScalePointFeature(u.2) Machine Vision Toolbox for MATLAB R 84 Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES Properties u horizontal coordinate v vertical coordinate strength feature strength scale feature scale descriptor feature descriptor (vector) Properties of a vector of ScalePointFeature objects are returned as a vector. A Indicate scale by a circle (default) Indicate scale by a translucent disk Color of circle or disk (default green) Transparency of disk. scale) as above but with speciﬁed feature scale.plot scale Plot feature scale F. SiftPointFeature ScalePointFeature. strength.plot scale(options.u is a 2 × N matrix with each column the corresponding point coordinate. f = ScalePointFeature(u. Options ‘circle’ ‘disk’ ‘color’. F. If F is a vector (N × 1) of ScalePointFeature objects then F. f = ScalePointFeature(u. v. v) is a point feature object with speciﬁed coordinates. v.

The elements of the cell array are the point features for the corresponding element of the image sequence. A complete history of all tracks is maintained. R ‘nslots’. C. M Search radius for feature in next frame (default 20) Maximum number of tracks (default 800) Similarity threshold (default 0. Options ‘radius’. Machine Vision Toolbox for MATLAB R 85 Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES Tracker Track points in image sequence This class assigns each new feature a unique identiﬁer and tracks it from frame to frame until it is lost.Tracker Create new Tracker object T = Tracker(im. T ‘movie’. N ‘thresh’.CHAPTER 2. A vector of track history structures with elements id and uv which is the path of the feature.8) Write the frames as images into the folder M as with sequential ﬁlenames. See also PointFeature Tracker. im (HxWxS) is an image sequence and C (S × 1) is a cell array of vectors of PointFeature subclass objects. one per active track. During operation the image sequence is animated and the point features are overlaid along with annotation giving the unique identiﬁer of the track. Methods plot tracklengths Plot all tracks Length of all tracks Properties track history A vector of structures. options) is a new tracker object.

Tracker.display() displays a compact human-readable string representation of the Tracker object Notes • This method is invoked implicitly at the command line when the result of an expression is a Tracker object and the command has no trailing semicolon.display Display value T. Tracker.char Tracker.plot Show feature trajectories T.char() is a compact string representation of the Tracker parameters and status.char Convert to string s = T. FUNCTIONS AND CLASSES See also PointFeature Tracker.tracklengths() is a vector containing the length of every track. Machine Vision Toolbox for MATLAB R 86 Copyright c Peter Corke 2011 .tracklengths Length of all tracks T.plot() overlays the tracks of all features on the current plot. See also Tracker.CHAPTER 2.

Machine Vision Toolbox for MATLAB R 87 Copyright c Peter Corke 2011 . and their characteristics is displayed. S Notes: Return image with uint8 pixels (default) Return image with ﬂoat pixels Return image with double precision pixels Return greyscale image Apply gamma correction with gamma=G Subsample the image by S in both directions. options) is a Video object that acquires images from the local video camera speciﬁed by the string camera. G ‘scale’. If camera is ‘?’ a list of available cameras. S ‘resolution’. AxisWebCamera. FUNCTIONS AND CLASSES Video Class to read from local video camera A concrete subclass of ImageSource that acquires images from a local camera. Options ‘uint8’ ‘ﬂoat’ ‘double’ ‘grey’ ‘gamma’. otherwise the result is not predictable. Movie Video. Methods grab size close char Aquire and return the next image Size of image Close the image source Convert the object parameters to human readable string See also ImageSource. • The speciﬁed ‘resolution’ must match one that the camera is capable of.CHAPTER 2.Video Video camera constructor v = Video(camera. Obtain an image of size S=[W H].

about x as above but this is the command rather than functional form See also whos Machine Vision Toolbox for MATLAB R 88 Copyright c Peter Corke 2011 .char Convert to string V. Notes • the function will block until the next frame is acquired.close() closes the connection to the camera.close Close the image source V. about Compact display of variable type about(x) displays a compact line that describes the class and dimensions of x.CHAPTER 2. FUNCTIONS AND CLASSES Video. Video.grab() acquires an image from the camera. Video.char() is a string representing the state of the camera object in human readable form.grab Acquire image from the camera im = V.

If th1 is a column vector. the second for right. and the right image is cyan. useful if the images were captured with a non-human stereo baseline or ﬁeld of view. and each is one of: ‘r’ red ‘g’ green ‘b’ green ‘c’ cyan ‘m’ magenta a = anaglyph(left. color) as above but the string color describes the color coding as a string with 2 letters. See also stdisp angdiff Difference of two angles d = angdiff(th1. right) is an anaglyph image where the two images of a stereo pair are combined into a single image by coding them in two different colors. disp) as above but allows for disparity correction. Machine Vision Toolbox for MATLAB R 89 Copyright c Peter Corke 2011 . th2) returns the difference between angles th1 and th2 on the circle. The result is in the interval [-pi pi). color. d = angdiff(th) returns the equivalent angle to th in the interval [-pi pi). right. These adjustments are achieved by trimming the images. right. By default the left image is red. a = anaglyph(left. Return the equivalent angle in the interval [-pi pi). If disp is positive the disparity is increased. FUNCTIONS AND CLASSES anaglyph Convert stereo images to an anaglyph image a = anaglyph(left. the ﬁrst for left. if negative it is reduced. and th2 a scalar then return a column vector where th2 is modulo subtracted from the corresponding elements of th1. Use this option to make the images more natural/comfortable to view.CHAPTER 2.

If lambda is a column vector.boundary.y2). y1. 6500). For example: l = [380:10:700]’*1e-9. [x. Machine Vision Toolbox for MATLAB R 90 Copyright c Peter Corke 2011 . r2) is the correlation of the two boundary proﬁles R1 and r2.y1] and p2=[x2.y2]. See also RegionFeature. xcorr bresenham Generate a line p = bresenham(x1. Endpoints must be integer. FUNCTIONS AND CLASSES blackbody Compute blackbody emission spectrum E = blackbody(lambda. then E is a column vector whose elements correspond to to those in lambda. p = bresenham(p1. % visible spectrum e = blackbody(l. p2) as above but p1=[x1. % emission of sun plot(l. x2.y1) to (x2.CHAPTER 2. T) is the blackbody radiation power density [W/m3 ] at the wavelength lambda [m] and temperature T [K]. y2) is a list of integer coordinates for points lying on the line segement (x1. Each is an N × 1 vector of distances from the centroid of an object to points on its perimeter at equal angular increments. x is also N × 1 and is a correlation whose peak indicates the relative orientation of one proﬁle with respect to the other. r2) as above but also returns the relative scale s which is the size of object 2 with respect to object 1. e) boundmatch Match boundary proﬁles x = boundmatch(R1.s] = boundmatch(R1.

d is a table of points with rows of the form [X Y Z U V] where (X. See also CentralCamera ccdresponse CCD spectral response R = ccdresponse(lambda) is the spectral response of a typical silicon imaging sensor at the wavelength lambda.V] is the corresponding image plane coordinate. Reference An ancient Fairchild data book for a sensor with no IR ﬁlter ﬁtted. Notes: • This method cannot handle lense distortion.Z) is the coordinate of a world point and [U. [C. The response is normalized in the range 0 to 1.E] = camcald(d) as above but E is the maximum residual error after back substitution [pixels].CHAPTER 2.Y. If lambda is a vector then R is a vector of the same length whose elements are the response at the corresponding element of lambda. FUNCTIONS AND CLASSES See also icanvas camcald Camera calibration from data points C = camcald(d) is the camera matrix (3 × 4) determined by least squares from corresponding world and image-plane points. Machine Vision Toolbox for MATLAB R 91 Copyright c Peter Corke 2011 .

R. k = closest(a. Options ‘n’. and x is N × 3. that is. opt) return an N × 2 matrix whose rows deﬁne the coordinates [x. circle Compute points on a circle circle(C.CHAPTER 2. b) is the correspondence for N-dimensional point sets a (N × N A) and b (N × N B). but the circle is always in the xy-plane with a z-coordinate of C(3). b) as above and d1(I)=—a(I)-b(J)— is the distance of the closest point. R. R Machine Vision Toolbox for MATLAB 92 Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES See also rluminos cie primaries Deﬁne CIE primary colors p = CIE PRIMARIES() is a 3-vector with the wavelengths [m] of the CIE 1976 red. N Specify the number of points (default 50) closest Find closest points in N-dimensional space. C is normally 2 × 1 but if 3 × 1 then the circle is embedded in 3D.d1] = closest(a. x = circle(C. that the I’th column of a is closest to the Jth column of b.y] of points around the circumferance of a circle centred at C and of radius R. [k. green and blue primaries respectively. opt) plot a circle centred at C with radius R. k (1 x NA) is such that the element J = k(I).

and 22500 (444.uk • The Stiles & Burch 2-deg CMFs are based on measurements made on 10 observers.) • These CMFs differ slightly from those of Stiles & Burch (1955).3) of Wyszecki & Stiles (1982) is gives them in interpolated 1 nm steps. (Table 1(5. Notes • Is a MEX ﬁle. See also distance cmfrgb Color matching function Tthe color matching function is the tristimulus required to match a particular wavelength excitation. rgb = cmfrgb(lambda) is the CIE color matching function (N × 3) for illumination at wavelength lambda (N × 1) [m]. while Table I(5. unlike the CIE 2 deg functions (which were reconstructed from chromaticity data). The data are referred to as pilot data.5.ioo.ac.32).3) of Wyszecki & Stiles (1982) gives the Stiles & Burch functions in 250 cm-1 steps. b) as above but also returns the distance to the second closest point. since. • From Table I(5. If lambda is a vector then each row of rgb is the color matching function of the corresponding element of lambda. Notes • Data from http://cvrl.44) cm-1 Machine Vision Toolbox for MATLAB R 93 Copyright c Peter Corke 2011 .CHAPTER 2.5.3) of Wyszecki & Stiles (1982).16). 19000 (526.ucl.5.d1.d2] = closest(a.5. As noted in footnote a on p. the CMFs have been ”corrected in accordance with instructions given by Stiles & Burch (1959)” and renormalized to primaries at 15500 (645.3) of Wyszecki & Stiles (1982). they were measured directly. FUNCTIONS AND CLASSES [k. E) is the CIE color matching (1×3) function for an illumination spectrum deﬁned by intensity E (N × 1) and wavelength lambda (N × 1) [m]. rgb = cmfrgb(lambda. but probably represent the best estimate of the 2 deg CMFs. 335 of Table 1(5.

Note • CIE 1931 2-deg xyz CMFs from cvrl.M). ccxyz col2im Convert pixel vector to image out = col2im(pix. imsize is a 2-vector (N. Notes • The number of rows in pix must match the product of the elements of imsize. R Machine Vision Toolbox for MATLAB 94 Copyright c Peter Corke 2011 . If lambda is a vector then each row of xyz is the color matching function of the corresponding element of lambda.ac.CHAPTER 2. xyz = cmfxyz(lambda.uk See also cmfrgb. FUNCTIONS AND CLASSES See also ccxyz cmfxyz matching function The color matching function is the XYZ tristimulus required to match a particular wavelength excitation. imsize) is an image (HxWxP) comprising the pixel values in pix (N × P ) with one row per pixel where N=H × W .ucl. xyz = cmfxyz(lambda) is the CIE xyz color matching function (N ×3) for illumination at wavelength lambda (N × 1) [m].ioo. im) as above but the dimensions of out are the same as im. E) is the CIE xyz color matching (1 × 3) function for an illumination spectrum deﬁned by intensity E (N × 1) and wavelength lambda (N × 1) [m]. out = col2im(pix.

rg) is the Euclidean distance on the rg-chromaticity plane from coordinate rg=[r. See also colorspace Machine Vision Toolbox for MATLAB R 95 Copyright c Peter Corke 2011 .g] to every pixel in the color image im. Notes • The output image could be thresholded to determine color similarity. FUNCTIONS AND CLASSES See also im2col colnorm Column-wise norm of a matrix cn = colnorm(a) returns an M × 1 vector of the normals of each column of the matrix a which is N × M . d is an image with the same dimensions as im and the value of each pixel is the color space distance of the corresponding pixel in im. • Note that Euclidean distance in the rg-chromaticity space does not correspond well with human perception of color differences. Perceptually uniform spaces such as Lab remedy this problem.CHAPTER 2. colordistance Colorspace distance d = colordistance(im.

color) is a color image where each pixel in out is set to the corresponding element of the greyscale image im or a speciﬁed color according to whether the corresponding value of mask is true or false respectively. @isnan. The color is speciﬁed as a 3-vector (R. im<100. options) is a segmentation of the color image im into k classes. A k-means clustering of the chromaticity of all input pixels is performed.G. mask.B). [1 0 0]) Notes • With no output arguments the image is displayed. ipixswitch colorkmeans Color image segmentation by clustering L = colorkmeans(im. The label image L has the same row and column dimension as im and each pixel has a value in the range 0 to k-1 which indicates which cluster the corresponding pixel belongs to. [L. and returns a per-pixel logical result. color) as above but a the mask is the return value of the function handle func applied to the image im. func.C] = colorkmeans(im. FUNCTIONS AND CLASSES colorize Colorize a greyscale image out = colorize(im. [0 0 1]) Display image with NaN values shown in red out = colorize(im. See also imono. k. @isnan. icolor. out = colorize(im. A k-means clustering of the chromaticity of all input pixels is performed.CHAPTER 2. Examples Display image with values < 100 in blue out = colorize(im. eg. k) as above but also returns the cluster centres C (k × 2) where the I’th row is the rg-chromaticity of the I’th cluster and corresponds to the label I. R Machine Vision Toolbox for MATLAB 96 Copyright c Peter Corke 2011 .

name = colorname(rgb) is a string giving the name of the color that is closest (Euclidean) to the given rgb-tristimulus value. the root mean square error of all pixel chromaticities with respect to their cluster centre. C) is a segmentation of the color image im into k classes which are deﬁned by the cluster centres C (k × 2) in chromaticity space.R] = colorkmeans(im. name = colorname(XYZ. Since cluster centres are provided the k-means segmentation step is not required. Options Various options are possible to choose the initial cluster centres for k-means: ‘random’ ‘spread’ ‘pick’ randomly choose k points from randomly choose k values within the rectangle spanned by the input chromaticities. “?burnt” Machine Vision Toolbox for MATLAB R 97 Copyright c Peter Corke 2011 . interactively pick cluster centres Notes • The k-means clustering algorithm used in the ﬁrst three forms is computationally expensive and time consuming. • The residual is an indication of quality of ﬁt. eg. • Clustering is performed in rg-chromaticity space. XYZ = colorname(name.C. FUNCTIONS AND CLASSES [L. L = colorkmeans(im. k) as above but also returns the residual R. See also colorname Map between color names and RGB values rgb = colorname(name) is the rgb-tristimulus value corresponding to the color speciﬁed by the string name. ‘xy’) is the XYZ-tristimulus value corresponding to the color speciﬁed by the string name. Notes • Color name may contain a wildcard. ‘xy’) is a string giving the name of the color that is closest (Euclidean) to the given XYZ-tristimulus value.CHAPTER 2. low is good. Pixels are assigned to the closest (Euclidean) centre.

CHAPTER 2. or alternatively. [o1. • MATLAB uses two standard data formats for R’G’B’: double data with intensities in the range 0 to 1. i1. s = ‘dest<-src’. and uint8 data with integer-valued intensities from 0 to 255. • Color space names are case insensitive.601) + Chroma Luma + Chroma (“digitized” version of Y’PbPr) NTSC PAL Y’UV Luma + Chroma NTSC Y’IQ Luma + Chroma SECAM Y’DbDr Luma + Chroma JPEG-Y’CbCr Luma + Chroma Hue Saturation Value/Brightness Hue Saturation Luminance/Intensity CIE XYZ CIE L*a*b* (CIELAB) CIE L*u*v* (CIELUV) CIE L*ch (CIELCH) Notes • RGB input is assumed to be gamma encoded • RGB output is gamma encoded • All conversions assume 2 degree observer and D65 illuminant. FUNCTIONS AND CLASSES • Based on the standard X11 color database rgb.i2. im) converts the image im to a different color space according to the string s which speciﬁes the source and destination color spaces. s = ‘src->dest’. • Tristimulus values are in the range 0 to 1 colorspace Color space conversion of image out = colorspace(s.o3] = colorspace(s. For example ‘yuv<’ is short for ‘yuv<-rgb’. Supported color spaces are: ‘RGB’ ‘YPbPr’ ‘YCbCr’/’YCC’ ‘YUV’ ‘YIQ’ ‘YDbDr’ ‘JPEGYCbCr’ ‘HSV’/’HSB’ ‘HSL’/’HLS’/’HSI’ ‘XYZ’ ‘Lab’ ‘Luv’ ‘Lch’ R’G’B’ Red Green Blue (ITU-R BT. im) as above but speciﬁes separate output channels or planes.txt.i3) as above but speciﬁes separate input channels. colorspace(s. As MATLAB’s native datatype. double data is the natural choice. • When R’G’B’ is the source or destination. it can be omitted.709 gamma-corrected) Luma (ITU-R BT. Input and output images have 3 planes.o2. and R Machine Vision Toolbox for MATLAB 98 Copyright c Peter Corke 2011 .

• If im is an M × 3 array. Given uint8 R’G’B’ color data. Notes • This fully vectorized (VERY FAST!) • It computes the Euclidean distance between two vectors by: ||A-B|| = sqrt ( ||A||ˆ2 + ||B||ˆ2 . for memory and computational performance. FUNCTIONS AND CLASSES the R’G’B’ format used by colorspace.J) is the distance between points a(I) and d(J). colorspace will ﬁrst cast it to double R’G’B’ before processing. out will also have size M × 3. B = rand(400. Author Pascal Getreuer 2005-2006 diff2 diff3(v) compute 2-point difference for each point in the vector v.b) is the Euclidean distances between L-dimensional points described by the matrices a (L × M ) and b (L × N ) respectively.100).200).B ) R Machine Vision Toolbox for MATLAB 99 Copyright c Peter Corke 2011 . distance Euclidean distances between sets of points d = distance(a. Example A = rand(400. some functions also operate with uint8 R’G’B’.CHAPTER 2. However.2*A. like a colormap.B). d = distance(A. The distance d is M × N and element d(I.

Tested: PC Matlab v5. non-zero is an object.3. tel. non zero is counter-clockwise.j). Thanx: Nikos Vlassis.2 and Solaris Matlab v5.(+31)20-5257524. seed. bunschot@wins. each row is one edge point coordinate (x. See also closest e2h Euclidean to homogeneous edgelist Return list of edge pixels for region E = edgelist(im. FUNCTIONS AND CLASSES Author Roland Bunschoten.y). Note that direction is with respect to y-axis upward. not image frame. Kruislaan 403 1098 SJ Amsterdam. The result E is a matrix. in matrix coordinate frame. direction) returns the list of edge pixels as above. but the direction of edge following is speciﬁed. E = edgelist(im. direction == 0 (default) means clockwise. Intelligent Autonomous Systems (IAS) group. University of Amsterdam. Notes • im is a binary image where 0 is assumed to be background. seed) return the list of edge pixels of a region in the image im starting at edge coordinate seed (i.CHAPTER 2.uva. Machine Vision Toolbox for MATLAB 100 R Copyright c Peter Corke 2011 . • seed must be a point on the edge of the region. • The seed point is always the ﬁrst element of the returned edgelist.nl Last Rev: Oct 29 16:35:48 MET DST 1999.

Points are speciﬁed by the columns of p. p1. 1998. p2) is the distance of the points p2 (2 × M ) from the epipolar lines due to points p1 (2 × N ) where f (3 × 3) is a fundamental matrix relating the views containing image points p1 and p2. epiline(f. See also fmatrix. p. FUNCTIONS AND CLASSES See also ilabel epidist Distance of point from epipolar line d = epidist(f. d (N × M ) is the distance matrix where element d(i. Oct 27. epidist Machine Vision Toolbox for MATLAB 101 R Copyright c Peter Corke 2011 .R. ls) as above but return a vector of graphic handles. Nuno Alexandre Cid Martins. fmatrix epiline Draw epipolar lines epiline(f. H = epiline(f.CHAPTER 2.S. See also epiline. I. Author Based on fmatrix code by. ls) as above but draw lines using the line style arguments ls. one per line drawn.j) is the distance from the point p2(j) to the epipolar line due to point p1(i). p. Coimbra. p) draws epipolar lines in current ﬁgure based on points p (2 × N ) and the fundamental matrix f (3 × 3).

and the convolution kernel has a half size of w. p2. that is. epidist gauss2d kernel k = gauss2d(im. Notes • The points must be corresponding. • f is a rank 2 matrix. Reference Hartley and Zisserman. Machine Vision Toolbox for MATLAB 102 R Copyright c Peter Corke 2011 . it is singular.CHAPTER 2.au/. epiline. sigma) Returns a unit volume Gaussian smoothing kernel. ‘Multiple View Geometry in Computer Vision’. Author Based on fundamental matrix code by Peter Kovesi. options) is the fundamental matrix (3 × 3) that relates two sets of corresponding points p1 (2 × N ) and p2 (2 × N ) from two different camera views. School of Computer Science & Software Engineering. The Gaussian has a standard deviation of sigma.uwa. k is (2W+1) x (2W+1). http://www. that is. • Contains a RANSAC driver. c. page 270. See also ransac. no outlier rejection is performed. which means it can be passed to ransac().edu.csse. The University of Western Australia. If w is not speciﬁed it defaults to 2*sigma. FUNCTIONS AND CLASSES fmatrix Estimate fundamental matrix f = fmatrix(p1. homography.

itriplepoint.y2). ithin. Unlike standard morphological operations S has three possible values: 0. See also plot homline Machine Vision Toolbox for MATLAB 103 R Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES h2e Homogeneous to Euclidean hitormiss Hit or miss transform H = hitormiss(im.CHAPTER 2. Homogeneous points X (3 × 1) on the line must satisfy L’*X = 0. 1 and don’t care (represented by NaN). iendpoint homline Homogeneous line from two points L = homline(x1. See also imorph.y1) and (x2. se) is the hit-or-miss transform of the binary image im with the structuring element se. x2. y2) returns a 3 × 1 vectors which describes a line in homogeneous form that contains the two Euclidean points (x1. y1.

no outlier rejection is performed. School of Computer Science & Software Engineering. See also ransac.uwa. Notes • The points must be corresponding.au/.csse.CHAPTER 2. p) applies homogeneous transformation T to the points stored columnwise in p. invhomog. which means it can be passed to ransac(). FUNCTIONS AND CLASSES homography Estimate homography H = homography(p1. http://www. fmatrix homtrans Apply a homogeneous transformation p2 = homtrans(T. Author Based on homography code by Peter Kovesi. • If T is in SE(2) (3 × 3) and – p is 2 × N (2D points) they are considered Euclidean (R2 ) – p is 3 × N (2D points) they are considered projective (p2 ) • If T is in SE(3) (4 × 4) and – p is 3 × N (3D points) they are considered Euclidean (R3 ) – p is 4 × N (3D points) they are considered projective (p3 ) Machine Vision Toolbox for MATLAB 104 R Copyright c Peter Corke 2011 . p2) is the homography (3 × 3) that relates two sets of corresponding points p1 (2×N ) and p2 (2×N ) from two different camera views of a planar object.edu. • The points must be projections of points lying on a world plane • Contains a RANSAC driver. The University of Western Australia.

but its position with respect to the input image is given by the second return value offs. See also e2h. if T = N × N and T=NxNxP then the result is NxNxP. S output image contains all the warped pixels.H] Notes • The edges of the resulting output image will in general not be be vertical and horizontal lines. R ‘scale’. If T1 is a 3-dimensional transformation then T is applied to each plane as deﬁned by the ﬁrst two dimensions. options) as above but offs is the offset of the warped tile out with respect to the origin of im. itrim. set unmapped pixels to this value output image contains the speciﬁed ROI in the input image scale the output by this factor ensure output image is D × D size of output image S=[W.CHAPTER 2. [out. options) is a warp of the image im obtained by applying the homography H to the coordinates of every input pixel. Options ‘full’ ‘extrapval’.offs] = homwarp(H. interp2 Machine Vision Toolbox for MATLAB 105 R Copyright c Peter Corke 2011 . that is tp=T*T1. ie. D ‘size’. See also homography. T1) applies homogeneous transformation T to the homogeneous transformation T1. FUNCTIONS AND CLASSES tp = homtrans(T. h2e homwarp Warp image by an homography out = homwarp(H. im. S ‘dimension’. im. V ‘roi’.

Animate image sequence with overlaid corner features: c = icorner(im. features (N × 1) cell array whose elements are vectors of feature objects. ’nfeat’. options) displays a greyscale image sequence im with point features overlaid. 179-187. Visual pattern recognition by moment invariants. options) displays a greyscale image sequence im (HxWxN) where N is the number of frames in the sequence. ’gs’). features.CHAPTER 2. See also npq ianimate Display an image sequence ianimate(im. The feature is plotted using the object’s plot method and additional options are passed through to that method. 1962. IRE Trans. features. IT-8:pp. % computer corners % features shown as green squares Machine Vision Toolbox for MATLAB 106 R Copyright c Peter Corke 2011 . Examples Animate image sequence: ianimate(seq). 200). Notes • im is assumed to be a binary image of a single connected region Reference M-K. FUNCTIONS AND CLASSES humoments Hu moments phi = humoments(im) is the vector (7 × 1) of Hu moment invariants for the binary image im. on Information Theory. Hu. ianimate(seq. ianimate(im.

CHAPTER 2. box = ibbox(im) is the minimal bounding box that contains the non-zero pixels in the image im. M ‘npoints’. options) is a vector of RegionFeature objects that describe each connected region in the image im. F ‘loop’ ‘movie’. idisp ibbox Find bounding box box = ibbox(p) is the minimal bounding box that contains the points described by the columns of p (2 × N ). iharris. YMIN YMAX]. N ‘only’. FUNCTIONS AND CLASSES Options ‘fps’. The bounding box is a 2 × 2 matrix [XMIN XMAX. isurf. Machine Vision Toolbox for MATLAB 107 R Copyright c Peter Corke 2011 . I set the frame rate (default 5 frames/sec) endlessly loop over the sequence save the animation as a series of PNG frames in the folder M plot no more than N features per frame (default 100) display only the I’th frame from the sequence See also PointFeature. iblobs features f = iblobs(im.

C set connectivity. vertical coordinate bounding box. [S1.0 ‘connect’. options) returns an edge image using the Canny edge detector.CHAPTER 2.0) 1 for a circle. horizontal coordinate centroid. A set pixel aspect ratio. 4 (default) or 8 ‘greyscale’ compute greyscale moments 0 (default) or 1 ‘boundary’ compute boundary (default off) ‘area’. imoments icanny edge detection E = icanny(im. minimum horizontal coordinate bounding box. The edges within im are marked by non-zero values in E. default 1. Machine Vision Toolbox for MATLAB 108 R Copyright c Peter Corke 2011 . maximum horizontal coordinate bounding box. ilabel. C accept only blobs of pixel value C (default all) The RegionFeature object has many properties including: uc vc umin umax vmin vmax area class label children edgepoint edge perimeter touch a b theta shape circularity moments centroid. less for other shapes a structure containing moments of order 0 to 2 See also RegionFeature. maximum vertical coordinate the number of pixels the value of the pixels forming this region the label assigned to this region a list of indices of features that are children of this feature coordinate of a point on the perimeter a list of edge points 2 × N matrix number of edge pixels true if region touches edge of the image major axis length of equivalent ellipse minor axis length of equivalent ellipse angle of major ellipse axis to horizontal axis aspect ratio b/a (always <= 1. and larger values correspond to stronger edges.S2] accept only blobs with shape in interval S1 to S2 ‘touch’ ignore blobs that touch the edge (default accept) ‘class’. minimum vertical coordinate bounding box.A2] accept only blobs with area in interval A1 to A2 ‘shape’. [A1. FUNCTIONS AND CLASSES Options ‘aspect’.

1 x strongest edge) set the upper hysteresis threshold (default 0. n) as above but the structuring element se is applied n times. se) is the image im after morphological closing with the structuring element se. that is n dilations followed by n erosions. See also iopen. This is an dilation followed by erosion. 1996-7. S ‘th0’. se. Tel Aviv University. imorph Machine Vision Toolbox for MATLAB 109 R Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES Options ‘sd’. kdgauss iclose closing out = iclose(im. T set the standard deviation for smoothing (default 1) set the lower hysteresis threshold (default 0.CHAPTER 2. the effective structuing element is the Minkowski sum of the structuring element with itself n times. out = iclose(im.5 x strongest edge) Author Oded Comay. T ‘th1’. See also isobel. Notes • Cheaper to apply a smaller structuring element multiple times than one large one.

ipixswitch iconcat Concatenate images C = iconcat(im. iconcat(im. C = icolor(im.options) as above but displays the concatenated images using idisp.CHAPTER 2. Examples Create a color image that looks the same as the greyscale image c = icolor(im). colorize. Create an aqua tinted version of the greyscale image c = icolor(im. The images do not have to be of the same size.u] = iconcat(im. [0 1 1]). See also imono. D ‘bgval’. [C. Options ‘dir’. and smaller images are surrounded by background pixels which can be speciﬁed. FUNCTIONS AND CLASSES icolor Colorize a greyscale image C = icolor(im) is a color image C (HxWx3)where each color plane is equal to im (H × W ). value of unset background pixels Machine Vision Toolbox for MATLAB 110 R Copyright c Peter Corke 2011 . color) as above but each output pixel is color (3 × 1) times the corresponding element of im.options) as above but also returns the vector u whose elements are the coordinates of the left (or top in vertical mode) edge of the corresponding image. B direction of concatenation: ‘horizontal’ (default) or ‘vertical’.options) concatenates images from the cell array im.

and contains only valid pixels Notes • This function is a convenience wrapper for the builtin function CONV2. options) convolves im1 with im2. The smaller image is taken as the kernel and convolved with the larger image. FUNCTIONS AND CLASSES Notes • Works for color or greyscale images • Direction can be abbreviated to ﬁrst character. ‘h’ or ‘v’ • In vertical mode all images are right justiﬁed • In horizontal mode all images are top justiﬁed See also idisp iconv Image convolution C = iconv(im1.CHAPTER 2. If the larger image is color (has multiple planes) the kernel is applied to each plane. Options ‘same’ ‘full’ ‘valid’ output image is same size as largest input image (default) output image is larger than the input image output image is smaller than the input image. resulting in an output image with the same number of planes. im2. See also conv2 Machine Vision Toolbox for MATLAB 111 R Copyright c Peter Corke 2011 .

‘noble’ or ‘klt’ kernel width for smoothing (default 2) kernel for gradient (default kdgauss(2)) set the value of k for Harris detector use a P × P patch of surrounding pixel values as the feature vector. This is a non-scale space detector and by default the Harris method is used. CM ‘cminthresh’. K ‘patch’.CHAPTER 2. The vector has zero mean and unit norm. P ‘color’ minimum corner strength minimum corner strength as a fraction of maximum corner strength don’t return features closer than E to the edge of image (default 2) don’t return a feature closer than R pixels to an earlier feature (default 0) return the N strongest corners (default Inf) choose the detector where D is one of ‘harris’ (default). specify that im is a color image not a sequence Notes • Corners are processed in order from strongest to weakest. D ‘sigma’. S ‘deriv’. • If im is NxMx3 it is taken as a sequence unless the option ‘color’ is given Machine Vision Toolbox for MATLAB 112 R Copyright c Peter Corke 2011 . D ‘k’. E ‘suppress’. R ‘nfeat’. FUNCTIONS AND CLASSES icorner Corner detector f = icorner(im. CT ‘edgegap’. options) is a vector of PointFeature objects describing detected corner features. The PointFeature object has many properties including: u v strength descriptor horizontal coordinate vertical coordinate corner strength corner descriptor (vector) Options ‘cmin’. • The function stops when: – the corner strength drops below cmin – the corner strenght drops below cMinThresh x strongest corner – the list of corners is exhausted • Features are returned in descending strength order • If im has more than 2 dimensions it is either a color image or a sequence • If im is NxMxP it is taken as an image sequence and f is a cell array whose elements are feature vectors for the corresponding image in the sequence. If im is an image sequence a cell array of PointFeature vectors is returned. N ‘detector’.

05 rad) stop after N iterations (default 100) stop when the relative change in error norm is less than T (default 0.5 [sec]. IEEE Computer Society. 593-593. FUNCTIONS AND CLASSES • If im is NxMx3xP it is taken as a sequence of color images and f is a cell array whose elements are feature vectors for the corresponding color image in the sequence. p2. vol. with a delay of 0. options) as above but also returns the norm of the error between the transformed point set p2 and p1. • The descriptor is a vector of ﬂoat types to save space References • “A combined corner and edge detector”. where * denotes squared and smoothed. N ‘mindelta’. isurf icp Point cloud alignment T = icp(p1. May 1988. with a delay of d [sec]. Harris and M. 1994. [T. pp.J.121-128. Manchester. C. • “Good features to track”.001) eliminate correspondences more than T x the median distance at each iteration. • The default descriptor is a vector [Ix* Iy* Ixy*] which are the unique elements of the structure tensor.6. pp. J. 1988. See also PointFeature. Stephens.d] = icp(p1. Tomasi. p2. Proc. T show the points p1 and p2 at each iteration. Fourth Alvey Vision Conf. d ‘plot’ ‘maxtheta’. J. limit the change in rotation at each step to T (default 0. Shi and C. pp 147-151. Machine Vision Toolbox for MATLAB 113 R Copyright c Peter Corke 2011 . options) is the homogeneous transformation that best transforms the set of points p2 to p1 using the iterative closest point algorithm.. show the points p1 and p2 at each iteration. Proc. • “Finding corners”.G. T ‘distthresh’. Image and Vision Computing.Noble. Options ‘dplot’. Computer Vision and Pattern Recognition.CHAPTER 2. T ‘maxiter’.

no. • For noisy data setting distthresh and maxtheta can help to prevent the the solution from diverging. sd) as above but the standard deviation of the smoothing kernel is set to sd. pp. idecimate an image s = idecimate(im.. Reference “A method for registration of 3D shapes”. IEEETrans. ismooth Machine Vision Toolbox for MATLAB 114 R Copyright c Peter Corke 2011 . 239-256. m. 2. []) as above but no smoothing is applied prior to decimation. • Smoothing is used to eliminate aliasing artifacts and the standard deviation should be chosen as a function of the maximum spatial frequency in the image. vol. Pattern Anal. 14. 1992. each plane is decimated. The image is smoothed with a Gaussian kernel with standard deviation m/2 then subsampled. See also iscale.or 3-dimensional. m.Besl and H. s = idecimate(im. Notes • If the image has multiple planes. Feb. FUNCTIONS AND CLASSES Notes • Points can be 2.McKay.CHAPTER 2. Intell. Mach. m) is a decimated version of the image im whose size is reduced by m (an integer) in both dimensions. P. s = idecimate(im.

Machine Vision Toolbox for MATLAB 115 R Copyright c Peter Corke 2011 . darker than ‘grey’. C ‘grey’ ‘invert’ ‘signed’ ‘invsigned’ ‘random’ ‘dark’ number of colors in the color map (default 256) display the image without the GUI no axes on the image no axes or frame on the image no axes. negative is red. positive is blue. If the image is zoomed.and y-axes respectively. negative is red. C ‘xydata’. • The “line” button allows two points to be speciﬁed and a new ﬁgure displays intensity along a line between those points. maximum value is white color map: greyscale unsigned. maximum value is black color map: greyscale signed. FUNCTIONS AND CLASSES idisp image display tool idisp(im. Options ‘ncolors’. options) displays an image and allows interactive investigation of pixel value. User interface: • Left clicking on a pixel will display its value in a box at the top. histogram and zooming. The image is displayed in a ﬁgure with a toolbar across the top. the histogram is computed over only those pixels in view. useful for displaying stereo pair display image planes (colors or sequence) as horizontally adjacent images y-axis increases upward. • The “histo” button displays a histogram of the pixel values in a new ﬁgure. N ‘nogui’ ‘noaxes’ ‘noframe’ ‘plain’ ‘bar’ ‘print’. zero is white color map: random values. they are ﬁrst concatenated (horizontally). image is inverted C is a 2-vector that speciﬁes the grey value range that spans the colormap. – The “zoom” button requires a left-click and drag to specify a box which deﬁnes the zoomed view. If im is a cell array of images. highlights ﬁne structure color map: greyscale unsigned. zero is white. zero is black color map: greyscale signed. zero is black. frame or GUI add a color bar to the image write the image to ﬁle F in EPS format display aspect ratio so that pixels are squate make ﬁgure very wide.CHAPTER 2. linear proﬁle. XY ‘colormap’. XY is a cell array whose elements are vectors that span the x. F ‘square’ ‘wide’ ‘ﬂatten’ ‘ynormal’ ‘cscale’. Set colormap C (N × 3) color map: greyscale unsigned. good for superimposed graphics Notes • Color images are displayed in true color mode: pixel triples map to display pixels • Grey scale images are displayed in indexed mode: the image pixel value is mapped through the color map to determine the display pixel value. positive is blue.

labels. idisplabel(im. iconcat idisplabel Display an image with mask idisplabel(im. colorseg idouble Convert integer image to double imd = idouble(im) returns an image with double precision elements in the range 0 to 1. The integer pixels are assumed to span the range 0 to the maximum value of their integer class. labelimage. See also iint Machine Vision Toolbox for MATLAB 116 R Copyright c Peter Corke 2011 . and labelimage is an N × M image containing integer pixel class labels for the corresponding pixels in im. icolorize. bg) as above but the grey level of the non-selected pixels is speciﬁed by bg in the range 0 to 1. caxis. The pixel classes to be displayed are given by the elements of labels which is a scalar a vector of class labels. labels) displays only those image pixels which belong to a speciﬁc class. which by default (’greyscale’) is the range black to white. Non-selected pixels are displayed as white. FUNCTIONS AND CLASSES • The minimum and maximum image values are mapped to the ﬁrst and last element of the color map. See also image.CHAPTER 2. colormap. labelimage. See also iblobs. im is a greyscale N × M or color NxMx3 image.

out = igamma(im. • For images of type int the pixels are assumed in the range 0 to max integer value. • For images with multiple planes the gamma correction is applied to all planes. Notes • Gamma encoding is typically performed in a camera with gamma=0. • For images of type double the pixels are assumed to be in the range 0 to 1.2.CHAPTER 2. hitormiss igamma correction out = igamma(im. Machine Vision Toolbox for MATLAB 117 R Copyright c Peter Corke 2011 . All pixels are raised to the power gamma. ithin. Computed using the hit-or-miss morphological operator. • Gamma decoding is typically performed in the display with gamma=2. gamma) is a gamma corrected version of im. See also itriplepoint. FUNCTIONS AND CLASSES iendpoint Find end points in a binary skeleton image out = iendpoint(im) is a binary image where pixels are set if the corresponding pixel in the binary image im is the end point of a single-pixel wide line such as found in an image skeleton. Gamma encoding can be performed with gamma > 1 and decoding with gamma < 1.45. ‘sRGB’) is a gamma decoded version of im using the sRGB decoding function (JPEG images sRGB encoded).

100. k. [l. k is the scale parameter. 2004.jpg’). L = igraphseg(im. Felzenszwalb and D. Int. min is the minimum region size (pixels). 167181. Example im = iread(’58060. Notes • Is a MEX ﬁle Author Pedro Felzenszwalb. vol. Sept. min) is a graph-based segmentation of the greyscale or color image im.CHAPTER 2. min. pp. imser Machine Vision Toolbox for MATLAB 118 R Copyright c Peter Corke 2011 .m] = igraphseg(im. idisp(im) Reference “Efﬁcient graph-based image segmentation”. Huttenlocher. k. sigma) as above but m is the number of regions found.5).5). 59. and a larger value indicates a preference for larger regions. sigma) as above and sigma is the width of a Gaussian which is used to initially smooth the image (default 0. min. P. 2006.m] = igraphseg(im. Journal on Computer Vision. L is an image of the same size as im where each element is the label assigned to the corresponding pixel in im. FUNCTIONS AND CLASSES igraphseg Graph-based segmentation L = igraphseg(im. k. 1500. 0. [L. See also ithresh.

FUNCTIONS AND CLASSES ihist Image histogram ihist(im.CHAPTER 2. For an image with multiple planes the histogram of each plane is given in a separate subplot. plot(x.x] = ihist(im.h). bar(x. options) displays the image histogram. [H. • For a uint8 image the MEX function fhist is used See also hist Machine Vision Toolbox for MATLAB 119 R Copyright c Peter Corke 2011 .h). Notes • For a uint8 image the histogram spans the greylevel range 0-255 • For a ﬂoating point image the histogram spans the greylevel range 0-1 • For ﬂoating point images all NaN and Inf values are ﬁrst removed. options) is the image histogram as a column vector. Options ‘nbins’ ‘cdf’ ‘normcdf’ ‘sorted’ number of histogram bins (default 256) compute a cumulative histogram compute a normalized cumulative histogram histogram but with occurrence sorted in descending order Example [h. options) as above but also returns the bin coordinates as a column vectors.x] = ihist(im). [h. H = ihist(im. For an image with multiple planes H is a matrix with one column per image plane.x] = ihist(im. ’normcdf’).

m. A value of 0 indicates that the region has no single enclosing region. FUNCTIONS AND CLASSES iint Convert image to integer class out = iint(im) is an image with 8-bit unsigned integer elements in the range 0 to 255. Machine Vision Toolbox for MATLAB 120 R Copyright c Peter Corke 2011 .y2). x1. The value of parents(I) is the label of the parent or enclosing region of region I. for a binary image this means the region touches the edge of the image.parents] = ilabel(im) as above but also returns region hierarchy information. ii is a precomputed integral image. [L. See also idouble iisum Sum of integral image s = iisum(ii. y2. for a multilevel image it means that it touches more than one other region. same size as im. Region labels are in the range 1 to M.y1) and bottom-right (x2. x2) is the sum of pixels in the rectangular image region deﬁned by its top-left (x1. See also intgimage ilabel Label an image L = ilabel(im) performs connectivity analysis on the image im and returns a label image L. The ﬂoating point pixels values in im are assumed to span the range 0 to 1.CHAPTER 2. [L.m] = ilabel(im) as above but returns the value of the maximum label value. y1. where each pixel value represents the integer region label assigned to the corresponding pixel in im.

FUNCTIONS AND CLASSES [L. imoments iline Draw a line in an image out = iline(im. [L. otherwise it does not. The pixels on the line are set to 1.parents. 8).Y]. The value of class(I) is the value of the pixels that comprise region I. v) as above but the pixel on the line are set to v. p1.class] = ilabel(im) as above but also returns the class of pixels within each region. To use 8-way connectivity pass a second argument of 8. • The image can be binary or multi-level • Connectivity is performed using 4 nearest neighbours by default.maxlabel. ilabel(im.parents. • This is a “low level” function. Notes • Is a MEX ﬁle.class. eg. p2. iproﬁle. p1.maxlabel. See also iblobs. Notes • Uses the Bresenham algorith • Only works for greyscale images • The line looks jagged since no anti-aliasing is performed See also bresenham. If edge(I) is 1 then region I touches edge of the image. ipaste Machine Vision Toolbox for MATLAB 121 R Copyright c Peter Corke 2011 . p2) is a copy of the image im with a line drawn between the points p1 and p2. IBLOBS is a higher level interface.edge] = ilabel(im) as above but also returns the edge-touch status of each region. out = iline(im. each a 2-vector [X.CHAPTER 2.

CC] where (DX. [xm. w2. xmax.DY. See also col2im imatch Template matching xm = imatch(im1. and columns the vertical position. s) works as above but also returns a matrix of matching score values for each template position tested. More generally s is a 4-vector s=[xmin. y.y). im2.y). The pixels are in image column order and there are N=W × H rows. • im1 and im2 must be the same size.y) and CC is the similarity score (zero-mean normalized cross correlation) for the best match in the search region. Notes • Useful for tracking a template in an image sequence. • ZNCC matching is used. The return value is xm=[DX. ymin. s] % relative to (x. The template in im1 is centred at (x. y. x. FUNCTIONS AND CLASSES im2col Convert an image to pixel per row format out = im2col(im) returns the image (HxWxP) as a pixel vector (N ×P ) where each row is a pixel value (1 × P ).CHAPTER 2. s. im2. -s. a perfect match score is 1. If s is a scalar the search region is [-s.and y-offsets relative to (x. ymax] relative to (x. H. The rows correspond to horizontal positions of the template.y) and of size s.score] = imatch(im1.0 Machine Vision Toolbox for MATLAB 122 R Copyright c Peter Corke 2011 .DY) are the x.y) and its half-width is H. centred at (x. x. The template is searched for within im2 inside a rectangular region. • Is a MEX ﬁle. s) is the matching subimage of im1 (template) within the image im2.

v] = imeshgrid(size) as above but the domain is described size which is scalar size× size or a 2-vector [w H]. All pixels are equally weighted. v. [u. effectively a binary image.v] = imeshgrid(w.u) = v. FUNCTIONS AND CLASSES See also isimilarity imeshgrid Domain matrices for image [u. effectively a greyscale image.v] = imeshgrid(im) return matrices that describe the domain of image im and can be used for the evaluation of functions over the image.CHAPTER 2. f = imoments(u. The element u(v. f = imoments(u. w) as above but the pixels have weights given by the vector w. The RegionFeature object has many properties including: Machine Vision Toolbox for MATLAB 123 R Copyright c Peter Corke 2011 . H) as above but the domain is w × H.u) = u and v(v. See also meshgrid imoments Image moments f = imoments(im) is a RegionFeature object that describes the greyscale moments of the image im. [u. v) as above but the moments are computed from the pixel coordinates given as vectors u and v.

vertical coordinate the number of pixels major axis length of equivalent ellipse minor axis length of equivalent ellipse angle of major ellipse axis to horizontal axis aspect ratio b/a (always <= 1. or its area. Options ‘r601’ ‘r709’ ITU recommendation 601 (default) ITU recommendation 709 See also colorize. m01.0) a structure containing moments of order 0 to 2. m02. options) is a greyscale equivalent to the color image im. m11. colorspace Machine Vision Toolbox for MATLAB 124 R Copyright c Peter Corke 2011 . the elements are m00. FUNCTIONS AND CLASSES uc vc area a b theta shape moments centroid. m20. Notes • For a binary image the zeroth moment is the number of non-zero pixels. imoments imono Convert color image to monochrome out = imono(im. if connected blobs are required then the ILABEL function must be used ﬁrst. horizontal coordinate centroid. See also RegionFeature. ilabel. Different conversion functions are supported. m10. icolor.CHAPTER 2. • This function does not perform connectivity.

L is an image of the same size as im where each element is the label assigned to the corresponding pixel in im. pp. J. Chum. m. O.CHAPTER 2. op) is the image im after morphological processing with the operator op and structuring element se. 2004. Sept. [label. ’double’). vol. idisp(im) Reference “Robust wide-baseline stereo from maximally stable extremal regions”. se. options) is a greyscale segmentation of the image im based on maximally stable extremal regions. ’light’). options) as above but m is the number of regions found.m] = imser(im. Image and Vision Computing.png’. Pajdla. ’grey’.m] = imser(im. by Andrea Vedaldi and Brian Fulkerson • vl mser is a MEX ﬁle Machine Vision Toolbox for MATLAB 125 R Copyright c Peter Corke 2011 . Urban. FUNCTIONS AND CLASSES imorph Morphological neighbourhood processing out = imorph(im. imser Maximally stable extremal regions L = imser(im. Matas. 22. Notes • Is a wrapper for vl mser. part of VLFeat (vlfeat. 761767. Options ‘dark’ ‘light’ looking for dark features against a light background (default) looking for light features against a dark background Example im = iread(’castle_sign2. The labels [L. and T.org).

Examples Create integral images for sum of pixels over rectangular regions i = intimage(im). Integral images can be used for rapid computation of summations over rectangular regions. Create integral images for sum of pixel squared values over rectangular regions i = intimage(im. FUNCTIONS AND CLASSES See also ithresh.ˆ2). See also ihist intgimage Compute integral image out = intimage(im) is an integral image corresponding to im. igraphseg inormhist Histogram normalization out = inormhist(im) is a histogram normalized version of the image im. Notes • The histogram of the normalized image is approximately uniform.CHAPTER 2. Machine Vision Toolbox for MATLAB 126 R Copyright c Peter Corke 2011 . • Highlights image detail in dark areas of an image.

CHAPTER 2. n) is a padded version of the image im with a block of NaN values n pixels wide on the sides of im as speciﬁed by sides which is a string containing one or more of the characters: ‘t’ top ‘b’ bottom ‘l’ left ‘r’ right out = ipad(im. v) as above but pads with pixels of value v. sides. Notes • Cheaper to apply a smaller structuring element multiple times than one large one. the effective structuing element is the Minkowski sum of the structuring element with itself n times. FUNCTIONS AND CLASSES See also iisum iopen Morphological opening out = iopen(im. out = iopen(im. se) is the image im after morphological opening with the structuring element se. that is n erosions followed by n dilations. n) as above but the structuring element se is applied n times. sides. imorph ipad Pad an image with constants out = ipad(im. See also iclose. se. n. Machine Vision Toolbox for MATLAB 127 R Copyright c Peter Corke 2011 . This is an erosion followed by dilation.

im1. ipaste Paste an image into an image out = ipaste(im. 255) Notes • Not a tablet computer. otherwise p is the top-left corner (default) the coordinates of p start at zero.V]. p. im2) is an image where each pixel is selected from the corresponding pixel in im1 or im2 according to the corresponding values of mask. 20. 0) Add a band of white pixels 10 pixels wide on all sides of the image: ipad(im. Machine Vision Toolbox for MATLAB 128 R Copyright c Peter Corke 2011 .CHAPTER 2. If the element of mask is zero im1 is selected. Options ‘centre’ ‘zero’ ‘set’ ‘add’ ‘mean’ The pasted image is centred at p. im2. ’tblr’. FUNCTIONS AND CLASSES Examples Add a band of zero pixels 20 pixels high across the top of the image: ipad(im. ’t’. options) is the image im with the image im2 pasted in at the position p=[U. otherwise im2 is selected. 10. by default 1 is assumed im2 overwrites the pixels in im (default) im2 is added to the pixels in im im2 is set to the mean of pixel values in im2 and im See also iline ipixswitch Pixelwise image merge out = ipixswitch(mask.

v (N × P ) has one row for each point along the line and the row is the pixel value which will be a vector for a multi-plane image. iline ipyramid Pyramidal image decomposition out = ipyramid(im) is a pyramid decomposition of input image im using Gaussian smoothing with standard deviation of 1. [p. Notes • The Bresenham algorithm is used to ﬁnd points along the line. p2) as above but also returns the coordinates of the pixels for each point along the line.uv] = iproﬁle(im. p1.v) for the corresponding row of p. See also bresenham. p1. Each row of uv is the pixel coordinate (u.CHAPTER 2. p2) is a vector of pixel values extracted from the image im (HxWxP) between the points p1 (2 × 1) and p2 (2 × 1). out is a cell array of images each one having R Machine Vision Toolbox for MATLAB 129 Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES Notes • im1 and im2 must have the same number of rows and columns • if im1 and im2 are both greyscale then out is greyscale • if either of im1 and im2 are color then out is color See also colorize iproﬁle Extract pixels along a line v = iproﬁle(im.

se). se) is a rank ﬁltered version of im. out = ipyramid(im. im > irank(im.CHAPTER 2.5)). sigma) as above but the Gaussian standard deviation is sigma. op. se. ismooth irank Rank ﬁlter out = irank(im.2) = 0. Machine Vision Toolbox for MATLAB 130 R Copyright c Peter Corke 2011 . The pyramid is computed down to a nonhalvable image size. idecimate. out = imorph(image. The highest rank. out = imorph(image. FUNCTIONS AND CLASSES dimensions half that of the previous image. sigma. n) as above but only n levels of the pyramid are computed. order. edge) as above but the processing of edge pixels can be controlled. nbins) as above but the number of histogram bins can be speciﬁed. hence output image had reduced dimensions. is order=1. out = ipyramid(im. the maximum. Only pixels corresponding to non-zero elements of the structuring element se are ranked and the order’th value in rank becomes the corresponding output pixel value. ones(5. se. The value of edge is: ‘border’ ‘none’ ‘trim’ ‘wrap’ the border value is replicated (default) pixels beyond the border are not included in the window output is not computed for pixels whose window crosses the border. 3 × 3 non-local maximum: se = ones(3.3). the image is assumed to wrap around Examples 5 × 5 median ﬁlter: irank(im. op. 1. Notes • Works for greyscale images only. nbins. 12. See also iscalespace. se(2.

convert image to greyscale if it’s color using ITU rec 601 convert image to greyscale if it’s color using ITU rec 709 gamma value. If multiple ﬁles match a 3D or 4D image is returned where the last dimension is the number of images in the sequence. either numeric or ‘sRGB’ decimate image by R in both dimensions apply the region of interest R to each image. On subsequent calls the initial folder is as set on the last call. options) reads the speciﬁed ﬁle and returns a matrix. R return an image with 8-bit unsigned integer pixels in the range 0 to 255 return an image with single precision ﬂoating point pixels in the range 0 to 1. where R=[umin umax. FUNCTIONS AND CLASSES Notes • Is a MEX ﬁle. • A histogram method is used with nbins (default 256). Options ‘uint8’ ‘single’ ‘double’ ‘grey’ ‘grey 709’ ‘gamma’. im = iread(ﬁle. R ‘roi’. vmin vmax]. See also imorph. return an image with double precision ﬂoating point pixels in the range 0 to 1. G ‘reduce’. iwindow iread Read image from ﬁle im = iread() presents a ﬁle selection GUI from which the user can select an image ﬁle which is returned as 2D or 3D matrix. Notes • A greyscale image is returned as an H × W matrix • A color image is returned as an HxWx3 matrix • A greyscale image sequence is returned as an HxWxN matrix where N is the sequence length Machine Vision Toolbox for MATLAB 131 R Copyright c Peter Corke 2011 . Wildcards are allowed in ﬁle names.CHAPTER 2. ivar. If the path is relative it is searched for on Matlab search path.

imono. Notes • Color images are not supported. • The resulting images may have negative disparity. k) is an image where each pixel is replicated into a k × k tile.h2] = irectify(f. istereo.out2] = irectify(f. Machine Vision Toolbox for MATLAB 132 R Copyright c Peter Corke 2011 . homwarp.CHAPTER 2.h1. m. path irectify Rectify stereo image pair [out1. See also FeatureMatch. CentralCamera ireplicate Expand image out = ireplicate(im. m. im1. FUNCTIONS AND CLASSES • A color image sequence is returned as an HxWx3xN matrix where N is the sequence length See also idisp. im1. f (3 × 3) is the fundamental matrix relating the two views and m is a FeatureMatch object containing point correspondences between the images. igamma. im2) as above but also returns the homographies h1 and h2 that warp im1 to out1 and im2 to out2 respectively. imwrite.out2. im2) is a rectiﬁed pair of images corresponding to im1 and im2. [out1. If im is H × W the result is (KH)x(KW). Notes • The resulting image pair are epipolar aligned.

vmax].H] return central part of image.R) is a subimage of the image im described by the rectangle R=[umin. See also idisp. V ‘smooth’. vmin. [out. iline irotate Rotate image out = irotate(im. S ‘extrapval’.umax. FUNCTIONS AND CLASSES See also idecimate. S ‘crop’ ‘scale’.vmin vmax]. options) is a version of the image im that has been rotated about its centre. out = iroi(im) as above but the image is displayed and the user is prompted to adjust a rubber band box to select the region of interest. Options ‘outsize’.R] = iroi(im) as above but returns the selected region of interest R=[umin umax. angle. same size as im scale the image size by S (default 1) set background pixels to V (default 0) smooth image with Gaussian of standard deviation S Machine Vision Toolbox for MATLAB 133 R Copyright c Peter Corke 2011 . S set size of out to H × W where S=[W.CHAPTER 2. iscale iroi Extract region of interest out = iroi(im.

s=[] means no smoothing (default s=1) Machine Vision Toolbox for MATLAB 134 R Copyright c Peter Corke 2011 . • Counter-clockwise is a positive angle.H] set background pixels to V (default 0) smooth image with Gaussian of standard deviation s. s>1 makes the image larger. bias=0. bias) as above but bias controls which part of the image is cropped. out = isamesize(im1. s<1 makes it smaller. Options ‘outsize’.5 moves the crop window down or to the right. bias<0.5 is symmetric cropping. s set size of out to H × W where s=[W. See also iscale isamesize Automatic image trimming out = isamesize(im1. while bias>0. s ‘extrapval’.CHAPTER 2. See also iscale iscale Scale an image out = iscale(im. FUNCTIONS AND CLASSES Notes • Rotation is deﬁned with respect to a z-axis into the image. im2. im2) is an image derived from im1 that has the same dimensions as im2 which is achieved by cropping and scaling.5 moves the crop window up or to the left. s) is a version of im scaled in both directions by s which is a real scalar. V ‘smooth’.

The standard deviation of the smoothing Gaussian is sigma. Notes • The Laplacian is computed from the difference of adjacent Gaussians. and s (n × 1) is the vector of scales. n) as above but sigma=1. irotate iscalemax Scale space maxima f = iscalemax(L. Notes • Features are sorted into descending feature strength.CHAPTER 2. in space and scale. s (N × 1) is a vector of scale values corresponding to each plane of L. Machine Vision Toolbox for MATLAB 135 R Copyright c Peter Corke 2011 . corresponding to each step of the sequence. sigma) is a scale space image sequence of length n derived from im (H × W ). FUNCTIONS AND CLASSES See also ireplicate. At each scale step the variance of the Gaussian increases by sigma2 . n. The ﬁrst step in the sequence is the original image. idecimate. [g.L. ScalePointFeature iscalespace Scale-space image sequence [g. s) is a vector of ScalePointFeature objects which are the maxima. of the Laplacian of Gaussian (LoG) scale-space image sequence L (HxWxN). g (HxWxN) is the scale sequence.L.s] = iscalespace(im. L (HxWxN) is the absolute value of the Laplacian of Gaussian (LoG) of the scale sequence.s] = iscalespace(im. See also iscalespace.

klog iscolor Test for color image iscolor(im) is true (1) if im is a color image.CHAPTER 2. ishomog(T. that is. options) returns a vector of SiftPointFeature objects representing scale and rotationally invariant interest points in the image im. isvec isift SIFT feature extractor sf = isift(im. The SiftPointFeature object has many properties including: Machine Vision Toolbox for MATLAB 136 R Copyright c Peter Corke 2011 . ilaplace. ‘valid’) as above. See also isrot. FUNCTIONS AND CLASSES See also iscalemax. ismooth. else false (0). but also checks the validity of the rotation matrix. it its third dimension is equal to three. ishomog Test if argument is a homogeneous transformation ishomog(T) is true (1) if the argument T is of dimension 4 × 4 or 4x4xN.

International Journal of Computer Vision. icorner isimilarity Locate template in image s = isimilarity(T. • The SIFT algorithm is patented by Univerity of British Columbia.CHAPTER 2. See also SiftPointFeature. Lowe. each of which is the feature vectors for the corresponding image in the sequence. • Wraps a MEX ﬁle from www. isurf. • Features are returned in descending strength order. FUNCTIONS AND CLASSES u v strength descriptor sigma theta horizontal coordinate vertical coordinate feature strength feature descriptor (128 × 1) feature scale feature orientation [rad] Options ‘nfeat’. Reference David G. 2 (2004). • ISURF is a functional equivalent.vlfeat. im) is an image where each pixel is the ZNCC similarity of the template T (M × M ) to the M × M neighbourhood surrounding the corresonding Machine Vision Toolbox for MATLAB 137 R Copyright c Peter Corke 2011 . 91-110. N ‘suppress’. pp.org • Corners are processed in order from strongest to weakest. R set the number of features to return (default Inf) set the suppression radius (default 0) Notes • Greyscale images only. “Distinctive image features from scale-invariant keypoints”. 60. • If im is HxWxN it is considered to be an image sequence and F is a cell array with N elements.

FUNCTIONS AND CLASSES input pixel in im. @ncc. • The ZNCC function is a MEX ﬁle and therefore the fastest • User provided similarity metrics can be provided. Notes • Similarity is not computed where the window crosses the image boundary. R Machine Vision Toolbox for MATLAB 138 Copyright c Peter Corke 2011 .p] = isize(im) is the image height H and width w and number of planes p. @ssd.d) is the size of the d’th dimension of im. metric) as above but the similarity metric is speciﬁed by the function metric which can be any of @sad. See also imatch.CHAPTER 2. ncc. @zsad. sigma) is the image im after convolution with a Gaussian kernel of standard deviation sigma.H] = isize(im) is the image height H and width w. zncc isize Size of image n = isize(im. See also size ismooth Gaussian smoothing out = ismooth(im. Even if the image has only two dimensions p will be one. [w. ssd. im. [w. the function accepts two regions and returns a scalar similarity score. s = isimilarity(T. zsad. zssd. s is same size as im.H. @zssd. and these output pixels are set to NaN. sad.

gy] = isobel(im. options) as above but the options are passed to CONV2. The Sobel kernel is: | -1 | -2 | -1 0 0 0 1| 2| 1| out = isobel(im. [gx.dx) as above but applies the kernel dx and dx’ to compute the horizontal and vertical gradients respectively. sigma. • Smooths all planes of the input image • The Gaussian kernel has a unit volume • If input image is integer it is converted to ﬂoat. See also iconv. Notes • By default (option ‘full’) the returned image is larger than the passed image. kgauss isobel Sobel edge detector out = isobel(im) is an edge image computed using the Sobel edge operator applied to the image im.CHAPTER 2.dx) as above but returns the gradient images. This is the norm of the vertical and horizontal gradients at each pixel. Machine Vision Toolbox for MATLAB 139 R Copyright c Peter Corke 2011 . those where the kernel does not exceed the bounds of the image. then converted back to integer. Options ‘full’ ‘same’ ‘valid’ returns the full 2-D convolution (default) returns out the same size as im returns the valid pixels only. FUNCTIONS AND CLASSES out = ismooth(im.gy] = isobel(im) as above but returns the gradient images. convolved. [gx.

isvec istereo Stereo matching d = istereo(iml. • The resulting image is the same size as the input image. options) as above but returns sim which is the same size as d and the elements are the peak matching score for the corresponding Machine Vision Toolbox for MATLAB 140 R Copyright c Peter Corke 2011 . range. imr. d (H × W ) is the disparity and the value at each pixel is the horizontal shift of the corresponding pixel in iml as observed in imr. the disparity d=d(v. imr.sim] = istereo(iml.M] for an N × M window. H is the half size of the matching window. See also ksobel. w. which can be a scalar for N × N or a 2-vector [N. ‘valid’) as above. range is the disparity search range. range. but also checks the validity of the rotation matrix. which can be a scalar for disparities in the range 0 to range. options) is a disparity image computed from the epipolar aligned stereo pair: the left image iml (H × W ) and the right image imr (H × W ). isrot(R.u) means that imr(v. iconv isrot Test if argument is a rotation matrix isrot(R) is true (1) if the argument is of dimension 3 × 3.u-d) is the same world point as iml(v.CHAPTER 2. or a 2-vector [DMIN DMAX] for searches in the range DMIN to DMAX. See also ishomog. icanny. H. That is.u). [d. else false (0). FUNCTIONS AND CLASSES Notes • Tends to produce quite thick edges.

w. 3) See also irectify. ‘ncc’.p] = istereo(iml. enable subpixel interpolation and d contains non-integer values (default false) Notes • Images must be greyscale.sim.5 to +0. imr. [d. In this case d is the interpolated disparity and p is a structure with elements A. For the default matching metric ZNCC this varies between -1 (very bad) to +1 (perfect). ‘ssd’ or ‘sad’. dx. Machine Vision Toolbox for MATLAB 141 R Copyright c Peter Corke 2011 . options) if the ‘interp’ option is given then disparity is estimated to sub-pixel precision using quadratic interpolation. B.A and p. M ‘interp’ string that speciﬁes the similarity metric to use which is one of ‘zncc’ (default). p.max) as above but pixel values lie in the range 0 to max. p. FUNCTIONS AND CLASSES elements of d. The interpolation polynomial is s = Ad2 + Bd + C where s is the similarity score and d is disparity relative to the integer disparity at which s is maximum.B are matrices the same size as d whose elements are the per pixel values of the interpolation polynomial coefﬁcients. • sim = max(dsi. stdisp istretch Image normalization out = istretch(im) is a normalized image in which all pixel values lie in the range 0 to 1.CHAPTER 2. range. a linear mapping where the minimum value of im is mapped to 0 and the maximum value of im is mapped to 1. range. options) as above but returns dsi which is the disparity space image (HxWxN) where N=DMAX-DMIN+1. • Disparity values pixels within a half-window dimension (H) of the edges will not be valid and are set to NaN. Options ‘metric’.dx is the peak of the polynomial with respect to the integer disparity at which s is maximum (in the range -0. w.dsi] = istereo(iml. [d. imr.sim.5). out = istretch(im. The I’th plane is the similarity of iml to imr shifted by DMIN+I-1. That is.

• Features are returned in descending strength order • If im is HxWxN it is considered to be an image sequence and F is a cell array with N elements. FUNCTIONS AND CLASSES See also inormhist isurf SURF feature extractor sf = isurf(im. • The sign of the Laplacian is not retained. Increasing the threshold reduces the number of features computed and reduces computation time. Twente) or a MEX-ﬁle OpenCV wrapper by Petter Strandmark. Machine Vision Toolbox for MATLAB 142 R Copyright c Peter Corke 2011 . R set the number of features to return (default Inf) set Hessian threshold. T ‘octaves’. number of octaves to process (default 5) return 128-element descriptor (default 64) don’t compute rotation invariance set the suppression radius (default 0). The SurfPointFeature object has many properties including: u v strength descriptor sigma theta horizontal coordinate vertical coordinate feature strength feature descriptor (64 × 1 or 128 × 1) feature scale feature orientation [rad] Options ‘nfeat’. Notes • Color images.CHAPTER 2. or sequences. N ‘thresh’. are ﬁrst converted to greyscale. N ‘extended’ ‘upright’ ‘suppress’. options) returns a vector of SurfPointFeature objects representing scale and rotationally invariant interest points in the image im. Kroon (U. each of which is the feature vectors for the corresponding image in the sequence. • Wraps an M-ﬁle implementation of OpenSurf by D. Features are not returned if they are within R [pixels] of an earlier (stronger) feature.

icorner isvec Test if argument is a vector isvec(v) is true (1) if the argument v is a 3-vector. Otherwise false (0). else false (0). “SURF: Speeded Up Robust Features”. L) is true (1) if the argument v is a vector of length L. Computer Vision and Image Understanding (CVIU). Tinne Tuytelaars. iendpoint Machine Vision Toolbox for MATLAB 143 R Copyright c Peter Corke 2011 . See also hitormiss. No. Luc Van Gool. isrot ithin Morphological skeletonization out = ithin(im) is the binary skeleton of the binary image im. Vol. pp. 3. isift. either a row. 346–359.or columnvector. isvec(v. 2008 See also SurfPointFeature.CHAPTER 2. See also ishomog. Any non-zero region is replaced by a network of single-pixel wide lines. 110.delay) as above but graphically displays each iteration of the skeletonization algorithm with a pause of delay seconds between each iteration. itriplepoint. FUNCTIONS AND CLASSES Reference Herbert Bay. Andreas Ess. out = ithin(im.

FUNCTIONS AND CLASSES ithresh Interactive image threshold ithresh(im) displays the image im in a window with a slider which adjusts the binary threshold. irectify Machine Vision Toolbox for MATLAB 144 R Copyright c Peter Corke 2011 .CHAPTER 2.out2] = itrim(im1. • For a uint8 class image the slider range is 0 to 255.im2) returns the central parts of images im1 and im2 as out1 and out2 respectively. See also homwarp. • For a ﬂoating point class image the slider range is 0 to 1.5. It assumes that the undeﬁned pixels in im1 and im2 have values of NaN.out2] = itrim(im1.im2. When images are rectiﬁed or warped the shapes can become quite distorted and are embedded in rectangular images surrounded by black of NaN values. T) as above but the initial threshold is set to T. [out1. Notes • Greyscale image only. This function crops out the central rectangular region of each. a lower value will include more.0 See also idisp itrim Trim images [out1. The default is 0.T) as above but the threshold T in the range 0 to 1 is used to adjust the level of cropping. ithresh(im. a higher value will include fewer NaN value in the result. The same cropping is applied to each input image.

se. ithin.CHAPTER 2. edge) as above but performance at edge pixels can be controlled. that is where three single-pixel wide line intersect. Computed using the hit-or-miss morphological operator. Machine Vision Toolbox for MATLAB 145 R Copyright c Peter Corke 2011 . The value of edge is: ‘border’ ‘none’ ‘trim’ ‘wrap’ the border value is replicated (default) pixels beyond the border are not included in the window output is not computed for pixels whose window crosses the border. op) is an image where each output pixel is the speciﬁed statistic over the pixel neighbourhood indicated by the structuring element se which should have odd side lengths. hence output image had reduced dimensions. These are the Voronoi points in an image skeleton. See also iendpoint. The operation op is one of: ‘var’ variance ‘kurt’ Kurtosis or peakiness of the distribution ‘skew’ skew or asymmetry of the distribution out = ivar(im. The elements in the neighbourhood corresponding to non-zero elements in se are packed into a vector on which the required statistic is computed. hitormiss ivar Pixel window statistics out = ivar(im. FUNCTIONS AND CLASSES itriplepoint Find triple points out = itriplepoint(im) is a binary image where pixels are set if the corresponding pixel in the binary image im is a triple point. se. the image is assumed to wrap around Notes • Is a MEX ﬁle. op.

se. @std). Compute the standard deviation over a 3 × 3 window: iwindow(im. se. • Is slow since the function func must be invoked once for every output pixel.CHAPTER 2. Notes • Is a MEX ﬁle. The neighbourhood is deﬁned by the size of the structuring element se which should have odd side lengths.3). @max). the image is assumed to wrap around Example Compute the maximum value over a 5 × 5 window: iwindow(im. The value of edge is: ‘border’ ‘none’ ‘trim’ ‘wrap’ the border value is replicated (default) pixels beyond the border are not included in the window output is not computed for pixels whose window crosses the border. func) is an image where each pixel is the result of applying the function func to a neighbourhood centred on the corresponding pixel in im. iwindow iwindow Generalized spatial operator out = iwindow(im. The return value becomes the corresponding pixel value in out. See also ivar. func. edge) as above but performance of edge pixels can be controlled. irank Machine Vision Toolbox for MATLAB 146 R Copyright c Peter Corke 2011 . hence output image had reduced dimensions. out = iwindow(image. FUNCTIONS AND CLASSES See also irank.5). ones(5. ones(3. The elements in the neighbourhood corresponding to non-zero elements in se are packed into a vector (in column order from top left) and passed to the speciﬁed function handle func.

s) as above but s is explicitly speciﬁed. Notes • If R is a 2-element vector the result is an annulus of ones. H) as above but the half-width is explictly speciﬁed. FUNCTIONS AND CLASSES kcircle Circular structuring element k = kcircle(R) is a square matrix S × S where S=2R+1 of zeros with a maximal centered circular region of radius R pixels set to one. and the two numbers are interpretted as inner and outer radii. dG/dy. k = kcircle(R.CHAPTER 2. Notes • This kernel is the horizontal derivative of the Gaussian. ktriangle. • This kernel is an effective edge detector. See also ones. dG/dx. See also kgauss. k = kdgauss(sigma. is k’. iconv Machine Vision Toolbox for MATLAB 147 R Copyright c Peter Corke 2011 . kdog. klog. imorph kdgauss Derivative of Gaussian kernel k = kdgauss(sigma) is a 2-dimensional derivative of Gaussian kernel (W × W ) of width (standard deviation) sigma and centred within the matrix k whose half-width H = 3 × sigma and W=2 × H+1. • The vertical derivative.

KGAUSS(SIGMA2). iconv kgauss Gaussian kernel k = kgauss(sigma) is a 2-dimensional unit-volume Gaussian kernel of width (standard deviation) sigma. kdog. The kernel is centred within the matrix k whose half-width H = 3 × SIGM A and W=2 × H+1. See also kdgauss. FUNCTIONS AND CLASSES kdog Difference of Gaussian kernel k = kdog(sigma1) is a 2-dimensional difference of Gaussian kernel equal to KGAUSS(sigma1) . and centred within the matrix k whose half-width is H=2 × sigma and W=2 × H+1. H) as above but the half-width H is speciﬁed.CHAPTER 2. iconv Machine Vision Toolbox for MATLAB 148 R Copyright c Peter Corke 2011 . sigma2) as above but sigma2 is speciﬁed directly. sigma2. k = kgauss(sigma. Notes • This kernel is similar to the Laplacian of Gaussian and is often used as an efﬁcient approximation. k = kdog(sigma1.6*sigma1. k = kdog(sigma1. H) as above but the kernel half-width is speciﬁed. where sigma1 > SIGMA2. klog. By default SIGMA2 = 1. klog. See also kgauss. kdgauss.

kdgauss. iconv. kdog. and W=2 × H+1. iconv klog Laplacian of Gaussian kernel k = klog(sigma) is a 2-dimensional Laplacian of Gaussian kernel of width (standard deviation) sigma and centred within the matrix k whose half-width is H=3 × sigma. k = klog(sigma. H) as above but the half-width H is speciﬁed.CHAPTER 2. FUNCTIONS AND CLASSES klaplace Laplacian kernel k = klaplace() is the Laplacian kernel: —0 |1 1 0— -4 1| —0 1 0— Notes • This kernel has an isotropic response to gradient. zcross Machine Vision Toolbox for MATLAB 149 R Copyright c Peter Corke 2011 . See also kgauss. See also ilaplace.

k. options) is k-means clustering of multi-dimensional data points x (D × N ) where N is the number of points. [L. FUNCTIONS AND CLASSES kmeans K-means clustering [L. it is assumed to have been completed previously. Reference Tou and Gonzalez.C] = kmeans(x. k. L is a vector (N ×1) whose elements indicates which cluster the corresponding element of x belongs to. C (D × k) contains the cluster centroids and L (N × 1) indicates which cluster the corresponding element of x is closest to. L = kmeans(x.CHAPTER 2. Options ‘random’ ‘spread’ initial cluster centres are chosen randomly from the set of data points x initial cluster centres are chosen randomly from within the hypercube spanned by x. Pattern Recognition Principles. The data is organized into k clusters based on Euclidean distance from cluster centres C (D × k). C) is similar to above but the clustering step is not performed. pp 94 ksobel Sobel edge detector k = ksobel() is the Sobel x-derivative kernel: |-1 |-2 |-1 0 0 0 1| 2| 1| Notes • This kernel is an effective horizontal edge detector • The Sobel vertical derivative is k’ Machine Vision Toolbox for MATLAB 150 R Copyright c Peter Corke 2011 .C] = kmeans(x. and D is the dimension. c0) as above but the initial clusters c0 (D × k) is given and column I is the initial estimate of the centre of cluster I.

CHAPTER 2. FUNCTIONS AND CLASSES

See also

isobel

ktriangle

Triangular kernel

k = ktriangle(w) is a triangular kernel within a rectangular matrix k. The dimensions k are w × w if w is scalar or w(1) wide and w(2) high. The triangle is isocles and is full width at the bottom row of the kernel and with its apex in the top row.

Examples

>> ktriangle(3) ans = |0 1 0| |0 1 0| |1 1 1|

See also

kcircle

lambda2rg

RGB chromaticity coordinates

rgb = lambda2rg(lambda) is the rg-chromaticity coordinate (1 × 2) for illumination at the speciﬁc wavelength lambda [metres]. If lambda is a vector (N × 1), then P (N × 2) is a vector whose elements are the luminosity at the corresponding elements of lambda. rgb = lambda2rg(lambda, E) is the rg-chromaticity coordinate (1 × 2) for an illumination spectrum E (N × 1) and lambda (N × 1).

Machine Vision Toolbox for MATLAB 151

R

Copyright c Peter Corke 2011

CHAPTER 2. FUNCTIONS AND CLASSES

See also

cmfrgb, lambda2xy

lambda2xy

= LAMBDA2XY(LAMBDA) is the xy-chromaticity coordinate (1 × 2) for

illumination at the speciﬁc wavelength LAMBDA [metres]. If LAMBDA is a vector (N × 1), then P (N × 2) is a vector whose elements are the luminosity at the corresponding elements of LAMBDA. xy = lambda2xy(lambda, E) is the rg-chromaticity coordinate (1 × 2) for an illumination spectrum E (N × 1) and lambda (N × 1).

See also

cmfxyz, lambda2rg

loadspectrum

Load spectrum data

s = loadspectrum(lambda, ﬁlename) is spectral data (N × D) from ﬁle ﬁlename interpolated to wavelengths [metres] speciﬁed in lambda (N × 1). The spectral data can be scalar (D=1) or vector (D>1) valued. [s,lambda] = loadspectrum(lambda, ﬁlename) as above but also returns the passed wavelength lambda.

Notes

• The ﬁle is assumed to have its ﬁrst column as wavelength in metres, the remainding columns are linearly interpolated and returned as columns of s.

Machine Vision Toolbox for MATLAB 152

R

Copyright c Peter Corke 2011

CHAPTER 2. FUNCTIONS AND CLASSES

luminos

Photopic luminosity function

p = luminos(lambda) is the photopic luminosity function for the wavelengths in lambda. If lambda is a vector (N × 1), then p (N × 2) is a vector whose elements are the luminosity at the corresponding elements of lambda. Luminosity has units of lumens which are the intensity with which wavelengths are perceived by the light-adapted human eye.

See also

rluminos

maxﬁlt

maximum ﬁlter

MAXFILT(s [,w]) minimum ﬁlter a signal with window of width w (default is 5) SEE ALSO: medﬁlt, minﬁlt pic 6/93

medﬁlt1

Median ﬁlter

y = medﬁlt1(x, w) is the one-dimensional median ﬁlter of the signal x computed over a sliding window of width w.

Notes

• A median ﬁlter performs smoothing but preserves sharp edges, unlike traditional smoothing ﬁlters.

Machine Vision Toolbox for MATLAB 153

R

Copyright c Peter Corke 2011

Machine Vision Toolbox for MATLAB 154 R Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES mkcube Create cube p = mkcube(s.z] = mkcube(s. By default the grid lies in the XY plane. If s is a 2-vector the side lengths are s(1)xS(2). C ‘T’. If d is a 2-vector the grid is d(1)xD(2) points.z] = mkcube(s. The cube is centred at C (3 × 1) not the origin The cube is arbitrarily transformed by the homogeneous transform T Return a set of cube edges in MATLAB mesh format rather than points.y. [x. options) is a set of points (3 x d2 ) that deﬁne a d × d planar grid of points with side length s. [x. Options ‘facepoint’ ‘centre’. options) as above but return the rows of p as three vectors. options) is a mesh that deﬁnes the edges of a cube. Options ‘T’. allowing the plane to be translated or rotated. in this case the returned value is 3 × 14 (8 vertices + 6 face centres). ‘edge’.CHAPTER 2.y. s. T ‘edge’ Add an extra point in the middle of each face. T the homogeneous transform T is applied to all points. sphere mkgrid Create grid of points p = mkgrid(d. options) is a set of points (3 × 8) that deﬁne the vertices of a cube of side length s and centred at the origin. The points are the columns of p. See also cylinder. symmetric about the origin.

FUNCTIONS AND CLASSES mlabel for mplot style graph mlabel(lab1 lab2 lab3) mplot multiple data Plot y versus t in multiple windows. p. n) MPLOT(t.CHAPTER 2. n is a row vector specifying which variables to plot (1 is ﬁrst data column. or y(:. the sum of I(x.yq . MPLOT(y) MPLOT(y. q) is the PQ’th moment of the image im. Subplots are labelled as per the data ﬁelds. {labels}) Where y is multicolumn data and t is time.xp . n. MPLOT(t. labels is a cell array of labels for the subplots. n is a row vector specifying which variables to plot (1 is ﬁrst data column. See also mpq poly. That is. {labels}) Where y is multicolumn data and ﬁrst column is time.y). mpq Image moments m = mpq(im.2)).2)). n) MPLOT(y. y. y. labels is a cell array of labels for the subplots. npq. MPLOT(S) Where S is a structure and one element ‘t’ is assumed to be time. y) MPLOT(t. or y(:. upq Machine Vision Toolbox for MATLAB 155 R Copyright c Peter Corke 2011 . Plot all other vectors versus time in subplots.

so centroids will be still be correct. Polygon mtools simple/useful tools to all windows in ﬁgure ncc Normalized cross correlation m = ncc(i1. • If the points are clockwise the moments will all be negated. q) is the PQ’th moment of the polygon with vertices described by the columns of v. The result m is a scalar in the interval -1 (non match) to 1 (perfect match) that indicates similarity.CHAPTER 2. FUNCTIONS AND CLASSES mpq poly Polygon moments m = MPQ POLY(v. i2) is the normalized cross-correlation between the two equally sized image patches i1 and i2. p. Notes • The points must be sorted such that they follow the perimeter in sequence (counterclockwise). Machine Vision Toolbox for MATLAB 156 R Copyright c Peter Corke 2011 . npq poly. • If the ﬁrst and last point in the list are the same. upq poly. they are considered to be a single vertex. See also mpq.

• w2 should be chosen to be half the “size” of the features to be segmented. 1986.2 Reference An Introduction to Digital Image Processing. -0.2. W. sad. See also zncc.s] = niblack(im. in text segmentation. idisp(im >= t). • The ncc similarity measure is invariant to scale changes in image intensity. The threshold at each pixel is a function of the mean and standard deviation computed over a W × W window. FUNCTIONS AND CLASSES Notes • A value of 1 indicates identical pixel patterns. ssd.m. where W=2*w2+1. niblack. Notes • This is an efﬁcient algorithm very well suited for binarizing text. • A common choice of k=-0. 20). k.CHAPTER 2. Example t = niblack(im. T has the same dimensions as im. the height of a character. Machine Vision Toolbox for MATLAB 157 R Copyright c Peter Corke 2011 . [T. Prentice-Hall. w2) is the per-pixel (local) threshold to apply to image im. k. isimilarity niblack Adaptive thresholding T = niblack(im. w2) as above but returns the per-pixel mean m and standard deviation s. for example.

p. b) npq Normalized central image moments m = npq(im.q)/MPQ(im. See also npq poly. p. Notes • The normalized central moments are invariant to translation and scale. upq npq poly Normalized central polygon moments m = NPQ POLY(v.0). That is UPQ(im. FUNCTIONS AND CLASSES See also otsu. q) is the PQ’th normalized central moment of the polygon with vertices described by the columns of v. mpq. Machine Vision Toolbox for MATLAB 158 R Copyright c Peter Corke 2011 . q) is the PQ’th normalized central moment of the image im.p.0. ithresh norm2 columnwise norm n = norm2(m) n = norm2(a.CHAPTER 2.

• If the ﬁrst and last point in the list are the same. FUNCTIONS AND CLASSES Notes • The points must be sorted such that they follow the perimeter in sequence (counterclockwise). • The normalized central moments are invariant to translation and scale. • If the points are clockwise the moments will all be negated. mpq. upq. npq. See also mpq poly. See also numrows numrows Return number of rows in matrix nr = numrows(m) returns the number of rows in the matrix m.CHAPTER 2. so centroids will be still be correct. they are considered as a single vertex. See also numcols Machine Vision Toolbox for MATLAB 159 R Copyright c Peter Corke 2011 . Polygon numcols Return number of columns in matrix nc = numcols(m) returns the number of columns in the matrix m.

otsu IEEE Trans. Man and Cybernetics Vol SMC-9(1). options) as above but also returns the indices of the maxima in the vector y. Notes • Performance for images with non-bimodal histograms can be quite poor. Reference A Threshold Selection Method from Gray-Level Histograms.i] = peak(y. Example t = otsu(im). Jan 1979. idisp(im >= t). x is the same length of y and contains the corresponding x-coordinates.xp] = peak(y. x. [yp. FUNCTIONS AND CLASSES otsu Threshold selection T = otsu(im) is an optimal threshold for binarizing an image with a bimodal intensity histogram. Systems. T is a scalar threshold that maximizes the variance between the classes of pixels below and above the thresold T.CHAPTER 2. N. Machine Vision Toolbox for MATLAB 160 R Copyright c Peter Corke 2011 . options) as above but also returns the corresponding x-coordinates of the maxima in the vector y. pp 62-66 See also niblack. ithresh peak Find peaks in vector yp = peak(y. options) are the values of the maxima in the vector y. [yp.

S ‘interp’ ‘plot’ Number of peaks to return (default all) Only consider as peaks the largest value in the horizontal and vertical range +/.ij] = peak2(z. N ‘scale’. Machine Vision Toolbox for MATLAB 161 R Copyright c Peter Corke 2011 . • The interp options ﬁts points in the neighbourhood about the peak with a paraboloid and its peak position is returned. options) as above but also returns the indices of the maxima in the matrix z. Order of interpolation polynomial (default no interpolation) Display the interpolation polynomial overlaid on the point data Notes • To ﬁnd minima. See also peak2 peak2 Find peaks in a matrix zp = peak2(z. N ‘scale’. FUNCTIONS AND CLASSES Options ‘npeaks’. Interpolate peak (default no interpolation) Display the interpolation polynomial overlaid on the point data Notes • To ﬁnd minima. use peak2(-V). N ‘plot’ Number of peaks to return (default all) Only consider as peaks the largest value in the horizontal range +/. S ‘interp’. options) are the peak values in the 2-dimensional signal z.S points.S points. • The interp options ﬁts points in the neighbourhood about the peak with an N’th order polynomial and its peak position is returned.CHAPTER 2. Typically choose N to be odd. Use SUB2IND to convert these to row and column coordinates Options ‘npeaks’. use peak(-V). [zp.

CHAPTER 2. See also pnmﬁlt. Notes • Provides access to a large number of Unix command line utilities such as ImageMagick. ie. Nx2xM or Nx3xM then the M trajectories are overlaid in the one plot. See also plot Machine Vision Toolbox for MATLAB 162 R Copyright c Peter Corke 2011 . ls) as above but the line style arguments ls are passed to plot. iread plot2 Plot trajectories plot2(p) plots a line with coordinates taken from successive rows of p. p can be N × 2 or N × 3. sub2ind pgmﬁlt Pipe image through PGM utility out = pgmﬁlt(im. If p has three dimensions. plot2(p. pgmcmd) pipes the image im through a Unix ﬁlter program and returns its output as an image. FUNCTIONS AND CLASSES See also peak. The program given by the string pgmcmd must accept and return images in PGM format.

’g’. ’b’). If C (2 × N or 3 × N ) and R (1 × N ) then a set of N circles are drawn with centre and radius taken from the columns of C and R. ‘g:’) for a non-ﬁlled circle.y1. plot_circle(c. and optional Matlab linestyle options ls. PLOT BOX(’centre’. 1=solid. 5). Options ‘edgecolor’ ‘ﬁllcolor’ ‘alpha’ the color of the circle’s edge. ‘size’. Machine Vision Toolbox for MATLAB 163 R Copyright c Peter Corke 2011 . Matlab color spec transparency of the ﬁlled circle: 0=transparent. W. ‘size’. plot_circle(c. YL YR] with optional Matlab linestyle options ls.y2. r. r. Notes • the option can be either a simple linespec (eg. ls) draws a box with corners at (x1. value pairs that are passed to plot. R. ’LineWidth’.Y] and with dimensions W=[WIDTH HEIGHT]. ’r’). PLOT BOX(x1. W. options) draws a circle on the current plot with centre C=[X Y] and radius R. r. ’edgecolor’. P. P. ‘r’. ls) draws a box deﬁned by b=[XL XR. FUNCTIONS AND CLASSES plot box a box on the current plot PLOT BOX(b. ls) draws a box with center at P=[X. ’fillcolor’. ls) draws a box with top-left at P=[X. PLOT BOX(’topleft’. plot circle Draw a circle on the current plot PLOT CIRCLE(C.y1) and (x2.y2). x2. or a set of name. If C=[X Y Z] the circle is drawn in the XY-plane at height Z.Y] and with dimensions W=[WIDTH HEIGHT]. Examples plot_circle(c. Matlab color spec the color of the circle’s interior.CHAPTER 2.

Y. plot frame Plot a coordinate frame represented by a homogeneous transformation trplot(T.Y]. FUNCTIONS AND CLASSES See also plot plot ellipse Draw an ellipse on the current plot PLOT ELLIPSE(a. current plot. with Matlab line style ls. plot ellipse inv Plot an ellipse plot ellipse(a. Matlab colorspec ‘axes’ ’axis’ Machine Vision Toolbox for MATLAB 164 R Copyright c Peter Corke 2011 . Options ‘color’. ls) as above but centred at C=[X.Z] the ellipse is parallel to the XY plane but at height Z. If C=[X.CHAPTER 2. centred at the origin. C. xc. options) draws a coordinate frame corresponding to the homogeneous transformation T. ls) draws an ellipse deﬁned by X’AX = 0 on the current plot. c Specify color of the axes. ls) ls is the standard line styles. PLOT ELLIPSE(a.

‘printf’. to ‘view’. ’r’). ‘framename’. ’B’) plot homline Draw a line in homogeneous form H = PLOT HOMLINE(L. ’name’. Machine Vision Toolbox for MATLAB 165 R Copyright c Peter Corke 2011 . Options ‘textcolor’. ls) draws a line in the current ﬁgure L. data Label points according to printf format string and corresponding element of data ‘sequence’ Label points sequentially Additional options are passed through to PLOT for creating the marker. ’color’. size Specify size of text ‘bold’ Text in bold font. ’r’. Matlab line speciﬁcation ls can be set. The current axis limits are used to determine the endpoints of the line. w ‘arrow’ ’length’. n Specify the name of the coordinate frame.CHAPTER 2. ’color’. trplot( T. See also homline plot point point features PLOT POINT(p.X = 0. l Specify length of the axes (default 1) Examples: trplot( T. The return argument is a vector of graphics handles for the lines. colspec Specify color of text ‘textsize’. n ‘text opts’. fmt. FUNCTIONS AND CLASSES ‘name’. v Specify the view angle for the Matlab axes ‘width’. where p is 2 × N and each column is the point coordinate. options) adds point markers to a plot.

• The number of vertices to draw the sphere is hardwired. 1=solid. See also plot. color) add spheres to the current ﬁgure. alpha) as above but alpha speciﬁes the opacity of the sphere were 0 is transparant and 1 is opaque. options ‘ﬁll’ ‘alpha’ the color of the circle’s interior. C is the centre of the sphere and if its a 3 × N matrix then N spheres are drawn with centres as per the columns. irrespective of ﬁgure hold state. color. Machine Vision Toolbox for MATLAB 166 R Copyright c Peter Corke 2011 . R.CHAPTER 2. R is the radius and color is a Matlab color spec. NOTES • The sphere is always added. H = PLOT SPHERE(C. FUNCTIONS AND CLASSES See also plot. patch. H = PLOT SPHERE(C. text plot poly Plot a polygon plotpoly(p. R. R. Polygon plot sphere Plot spheres PLOT SPHERE(C. color) as above but returns the handle(s) for the spheres. options) plot a polygon deﬁned by columns of p which can be 2 × N or 3 × N. The default is 1. Matlab color spec transparency of the ﬁlled circle: 0=transparent. either a letter or 3-vector.

plotp(p. p can be N × 2 or N × 3.CHAPTER 2. See also plot. pnmcmd) pipes the image im through a Unix ﬁlter program and returns its output as an image. FUNCTIONS AND CLASSES plotp Plot trajectories plotp(p) plots a set of points p. The program given by the string pnmcmd must accept and return images in PNM format. plot2 pnmﬁlt Pipe image through PNM utility out = pnmﬁlt(im. Notes • Provides access to a large number of Unix command line utilities such as ImageMagick. iread Machine Vision Toolbox for MATLAB 167 R Copyright c Peter Corke 2011 . ls) as above but the line style arguments ls are passed to plot. See also pgmﬁlt. which by Toolbox convention are stored one per column. By default a linestyle of ‘bx’ is used.

Notes • functions for T in SE(2) or SE(3) – if R is 2 × 2 then T is 3 × 3.CHAPTER 2. At each pixel the image gradient vector is resolved into the radial and tangential directions. See also isobel Machine Vision Toolbox for MATLAB 168 R Copyright c Peter Corke 2011 .gt] = radgrad(im. centre) as above but the centre of the image is speciﬁed as centre=[X. • translational component is zero See also t2r radgrad Radial gradient [gr. radgrad(im) as above but the result is displayed graphically. or – if R is 3 × 3 then T is 4 × 4. [gr.Y] rather than the centre pixel of im.gt] = radgrad(im) is the radial and tangential gradient of the image im. FUNCTIONS AND CLASSES r2t Convert rotation matrix to a homogeneous transform T = r2t(R) is a homogeneous transform equivalent to an orthonormal rotation matrix R with a zero translational component.

one column per point pair. T. d) as above but elements increment by d.in] = ransac(func. x. N maximum number of iterations (default 2000) maximum number of attempts to select a non-degenerate data set (default 100) Model function out = func(R) is the function passed to RANSAC and it must accept a single argument R which is a structure: Machine Vision Toolbox for MATLAB 169 R Copyright c Peter Corke 2011 . FUNCTIONS AND CLASSES ramp create a ramp vector ramp(n) output a vector of length n that ramps linearly from 0 to 1 ramp(v) as above but vector is same length as v ramp(v.resid] = ransac(func. N ‘maxDataTrials’. options) as above but returns the ﬁnal residual of applying func to the inlier set. Options ‘maxTrials’. [m. options) is the ransac algorithm that robustly ﬁts data x to the model represented by the function func. x.in. T. T is a threshold on how well a point ﬁts the estimated. ransac determines the subset of points (inliers) that best ﬁt the model described by the function func and the parameter m. T. if the ﬁt residual is aboe the the threshold the point is considered an outlier.CHAPTER 2. x typically contains corresponding point data. See also linspace ransac Random sample and consensus m = ransac(func. x. [m. ransac classiﬁes Points that support the model as inliers and those that do not as outliers. options) as above but returns the vector in of column indices of x that describe the inlier point set.

x = CONDITION(R. FUNCTIONS AND CLASSES R. they detect a structure argument. Boles. pp 381-395.valid is true if a set of points is not degenerate. 1981 • Richard Hartley and Andrew Zisserman.t threshold (1 × 1) R. fundamental matrix) it is necessary to condition the data to improve the accuracy of model estimation.x) condition the point data out.theta = DECONDITION(R. pp 101-113.x and returns the best model out.R. • The functions FMATRIX and HOMOG are written so as to be callable from RANSAC..out. For efﬁciency the data is conditioned once.cmd are: ‘size’ ‘condition’ ‘decondition’ ‘valid’ ‘estimate’ out.debug display what’s going on (logical) R. ”Random sample concensus: A paradigm for model ﬁtting with applications to image analysis and automated cartography”.x data to work on.inlier list of inliers (1 × m) out. Mach. [out.C.A. Assoc.theta. Comm. Comp. ”Multiple View Geometry in Computer Vision”. The inverse conditioning operation is applied to the model to transform the estimate based on conditioned data to a model applicable to the original data. 2001 Machine Vision Toolbox for MATLAB 170 R Copyright c Peter Corke 2011 .cmd the operation to perform which is either (string) R.out.x conditioned data (2D × N ) out. If this function cannot ﬁt a model then out.CHAPTER 2.x that best supports (most inliers) that model. This is used to discard random samples that do not result in useful models.resid model ﬁt residual (1 × 1) The values of R. ‘error’ Notes • For some algorithms (eg.theta] = ERR(R. N point pairs (6 × N ) R. that is.x) returns the best ﬁt model and residual for the subset of points R.s is the minimum number of points required to compute an estimate to out. Cambridge University Press. that is they will produce a model.inlier. References • m.resid] = EST(R.theta.x.x.theta to the points R.misc private data (cell array) out. If multiple models are found out.theta and the subset of R.s out.theta = []. [out.T) evaluates the distance from the model(s) R. No 6. Fishler and R.misc element.s sample size (1 × 1) out. Vol 24. and the data transform parameters are kept in the .theta) decondition the estimated model data out.valid if data is valid for estimation (logical) out.theta estimated quantity to test (3 × 3) R.theta is a cell array.theta estimated quantity (3 × 3) out.misc private data (cell array) The function return value is also a structure: out.

FUNCTIONS AND CLASSES Author Peter Kovesi School of Computer Science & Software Engineering The University of Western Australia pk at csse uwa edu au http://www.csse.edu. Relative luminosity lies in the interval 0 to 1 which indicate the intensity with which wavelengths are perceived by the light-adapted human eye. If lambda is a vector. homography rg addticks Label spectral locus RG ADDTICKS() adds wavelength ticks to the spectral locus.au/ pk See also fmatrix.CHAPTER 2. then p is a vector whose elements are the relative luminosity at the corresponding elements of lambda. See also luminos Machine Vision Toolbox for MATLAB 171 R Copyright c Peter Corke 2011 . See also xycolourspace rluminos Relative photopic luminosity function p = ruminos(lambda) is the relative photopic luminosity function for the wavelengths in lambda.uwa.

rotz. angvec2r roty Rotation about Y axis R = roty(theta) is a rotation matrix representing a rotation of theta about the y-axis. See also rotx. FUNCTIONS AND CLASSES rotx Rotation about X axis R = rotx(theta) is a rotation matrix representing a rotation of theta about the x-axis. See also roty.CHAPTER 2. roty. angvec2r rotz Rotation about Z axis R = rotz(theta) is a rotation matrix representing a rotation of theta about the z-axis. angvec2r Machine Vision Toolbox for MATLAB 172 R Copyright c Peter Corke 2011 . rotz. See also rotx.

yaw. pitch. If rpy has multiple rows they are assumed to represent a trajectory and R is a three dimensional matrix. • many texts (Paul. pitch and yaw are column vectors then they are assumed to represent a trajectory and R is a three dimensional matrix. where the last index corresponds to the rows of roll.CHAPTER 2. eul2tr rt2tr Convert rotation and translation to homogeneous transform TR = rt2tr(R. then TR is 3 × 3 – If R is 3 × 3 and t is 3 × 1. If roll. then TR is 4 × 4 • the validity of R is not checked Machine Vision Toolbox for MATLAB 173 R Copyright c Peter Corke 2011 . Notes • functions for R in SO(2) or SO(3) – If R is 2 × 2 and t is 2 × 1. pitch. See also tr2rpy. Spong) use the rotation order ZYX. Y. T = rpy2tr(roll. FUNCTIONS AND CLASSES rpy2tr Roll-pitch-yaw angles to homogeneous transform T = rpy2tr(rpy) is an orthonormal rotation matrix equivalent to the speciﬁed roll. yaw) as above but the roll-pitch-yaw angles are passed as separate arguments. yaw angles which correspond to rotations about the X. where the last index corresponds to the rows of rpy. pitch. Z axes respectively. t) is a homogeneous transformation matrix formed from an orthonormal rotation matrix R and a translation vector t. Note • in previous releases (<8) the angles corresponded to rotations about ZYX.

CHAPTER 2. tr2rt sad Sum of absolute differences m = sad(i1. FUNCTIONS AND CLASSES See also t2r. y.y] T = se2(xyt) as above where xyt=[x. a value of 0 indicates identical pixel patterns and is increasingly positive as image dissimilarity increases. r2t.theta] See also trplot2 Machine Vision Toolbox for MATLAB 174 R Copyright c Peter Corke 2011 . theta) is a 3 × 3 homogeneous transformation SE(2) representing translation x and y.y] and rotation is zero T = se2(xy. theta) as above where xy=[x. T = se2(xy) as above where xy=[x.y. and rotation theta in the plane. isimilarity se2 Create planar translation and rotation transformation T = se2(x. The result m is a scalar that indicates image similarity. ssd. ncc. i2) is the sum of absolute differences between the two equally sized image patches i1 and i2. See also zsad.

Clicking a point in the left image positions black cross hair at the same pixel coordinate in the right image. See also vex ssd Sum of squared differences m = ssd(i1. See also zsdd. Machine Vision Toolbox for MATLAB 175 R Copyright c Peter Corke 2011 . isimilarity stdisp Display stereo pair stdisp(L. a value of 0 indicates identical pixel patterns and is increasingly positive as image dissimilarity increases. ncc. FUNCTIONS AND CLASSES skew Create skew-symmetric matrix s = skew(v) is a skew-symmetric matrix and v is a 3-vector.CHAPTER 2. i2) is the sum of squared differences between the two equally sized image patches i1 and i2. sad. R) displays the stereo image pair L and R in adjacent windows. Clicking the corresponding world point in the right image sets the green crosshair and displays the disparity [pixels]. The result m is a scalar that indicates image similarity. Two cross-hairs are created.

– If T is 3 × 3.choose = {’this’. • the validity of rotational part is not checked See also r2t. opt. then R is 2 × 2. b. c. boolean or enumeration types (string or int).args] = TB OPTPARSE(opt. opt. It supports options that have an assigned value. ’that’. FUNCTIONS AND CLASSES See also idisp. istereo t2r Return rotational submatrix of a homogeneous transformation R = t2r(T) is the orthonormal rotation matrix component of homogeneous transformation matrix T: Notes • functions for T in SE(2) or SE(3) – If T is 4 × 4. varargin) opt. The software pattern is: function(a.foo = true.bar = false. ’other’}. tr2rt. then R is 3 × 3. opt. rt2tr tb optparse Standard option parser for Toolbox functions [optout. arglist) is a generalized option parser for Toolbox functions. Machine Vision Toolbox for MATLAB 176 R Copyright c Peter Corke 2011 .CHAPTER 2.blah = [].

If neither of ‘no’ or ‘yes’ are speciﬁed then opt. ’#yes’}.choose <.select <. The following options are automatically parsed: ‘verbose’ ‘debug’. If w is a scalar the image is w × w else w(2)xW(1).S displays opt and arglist testpattern Create test images im = testpattern(type.y ‘that’ sets opt.arglist] = tb_optparse(opt. The image is speciﬁed by the string type and one or two (type speciﬁc) arguments: Machine Vision Toolbox for MATLAB 177 R Copyright c Peter Corke 2011 . S ‘showopt’ sets opt.verbose <.false ‘blah’.x.y sets opt.CHAPTER 2. x.blah <.‘that’ ‘yes’ sets opt. varargin).blah <. FUNCTIONS AND CLASSES opt.true sets opt. Sometimes it is useful to collect the unassigned options and this can be achieved using a second output argument [opt.foo <. N ‘setopt’. If neither of ‘this’.1. The allowable options are speciﬁed by the names of the ﬁelds in the structure opt. ‘that’ or ‘other’ are speciﬁed then opt.foo <. w. By default if an option is given that is not a ﬁeld of opt an error is declared.N sets opt <. which is a cell array of all unassigned arguments in the order given in varargin.true ‘nobar’ sets opt.‘this’.select = {’#no’.choose <.select <. opt = tb_optparse(opt. varargin). Optional arguments to the function behave as follows: ‘foo’ sets opt.debug <.3 ‘blah’.2 (the second element) and can be given in any combination. 3 sets opt. if multiple values are required they must be converted to a cell array. Note: • that the enumerator names must be distinct from the ﬁeld names. The return structure is automatically populated with ﬁelds: verbose and debug. • that only one value can be assigned to a ﬁeld. args) creates a test pattern image.

CHAPTER 2. 2). FUNCTIONS AND CLASSES ‘rampx’ ‘rampy’ ‘sinx’ ‘siny’ ‘dots’ ‘squares’ ‘line’ intensity ramp from 0 to 1 in the x-direction. intercept. args is the number of cycles. args are pitch (distance between centres). binary dot pattern.sd. intensity ramp from 0 to 1 in the y-direction. args is the number of cycles. 50. 256. sinusoidal intensity pattern (from -1 to 1) in the y-direction. Notes • With no output argument the testpattern in displayed using idisp. sinusoidal intensity pattern (from -1 to 1) in the x-direction.sdd] = tpoly(s0. See also idisp tpoly Generate scalar polynomial trajectory [s. sd and sdd are n-vectors. args are theta (rad). sf. a line. binary square pattern. args is the number of cycles. A 256 × 256 image with a grid of dots on 50 pixel centres and 20 pixels in diameter: testpattern(’dots’. Velocity and acceleration can be optionally returned as sd and sdd. Examples A 256 × 256 image with 2 cycles of a horizontal sawtooth intensity ramp: testpattern(’rampx’. Machine Vision Toolbox for MATLAB 178 R Copyright c Peter Corke 2011 . args are dot pitch (distance between centres). square side length. dot diameter. 256.sdd] = tpoly(s0. [s.sd. args is the number of cycles. n) is a trajectory of a scalar that varies smoothly from s0 to sf in n steps using a quintic (5th order) polynomial. sf. 25). The trajectory s. T) as above but speciﬁes the trajectory in terms of the length of the time vector T.

v] = tr2angvec(T) is a rotation of theta about the vector v equivalent to the rotational component of the homogeneous transform T. The 3 angles rpy=[R.P. then each row of rpy corresponds to a step of the trajectory. [theta. See also angvec2r. If R or T represents a trajectory (has 3 dimensions).CHAPTER 2. rpy = tr2rpy(R. options) are the roll-pitch-yaw angles expressed as a row vector corresponding to the orthonormal rotation matrix R. angvec2tr tr2rpy Convert a homogeneous transform to roll-pitch-yaw angles rpy = tr2rpy(T. Y and Z axes respectively.Y] correspond to sequential rotations about the X. Machine Vision Toolbox for MATLAB 179 R Copyright c Peter Corke 2011 . options) are the roll-pitch-yaw angles expressed as a row vector corresponding to the rotation part of a homogeneous transform T. Notes • If no output arguments are speciﬁed the result is displayed. X axes (Paul book) Notes • There is a singularity for the case where THETA=0 in which case PHI is arbitrarily set to zero and PSI is the sum (PHI+PSI). FUNCTIONS AND CLASSES tr2angvec Convert rotation matrix to angle-vector form [theta. Y.v] = tr2angvec(R) is a rotation of theta about the vector v equivalent to the orthonormal rotation matrix R. Options ‘deg’ ‘zyx’ Compute angles in degrees (radians default) Return solution for sequential rotations about Z.

y.y.:.CHAPTER 2. See also rt2tr. then R is 2 × 2 and T is 2 × 1. If T has three dimensions. • The validity of R is not checked. Notes • Functions for TR in SE(2) or SE(3) – If TR is 4 × 4. p = transl(T) is the translational part of a homogenous transform as a 3-element column vector. 4x4xN then T is considered a homgoeneous Machine Vision Toolbox for MATLAB 180 R Copyright c Peter Corke 2011 .z]. ie. r2t. If p is an M × 3 matrix transl returns a 4x4xM matrix representing a sequence of homogenous transforms such that T(:. Spong) use the rotation order ZYX. T = transl(p) is a homogeneous transform representing a translation or point p=[x. See also rpy2tr. then R is 3 × 3 and T is 3 × 1. – If TR is 3 × 3.t] = tr2rt(TR) split a homogeneous transformation matrix into an orthonormal rotation matrix R and a translation vector t. tr2eul tr2rt Convert homogeneous transform to rotation and translation [R. z) is a homogeneous transform representing a pure translation. FUNCTIONS AND CLASSES • Note that textbooks (Paul. t2r transl Create translational transform T = transl(x.i) corresponds to the i’th row of p.

each N × 1.C2] = tristim2cc(tri) as above but the chromaticity coordinates are returned in separate vectors. or x and y. An historical anomaly. Machine Vision Toolbox for MATLAB 181 R Copyright c Peter Corke 2011 .o2] = tristim2cc(im) as above but the chromaticity is returned as separate images (H × W ). out (HxWx2) has planes corresponding to r and g.CHAPTER 2. [o1. out = tristim2cc(im) is the chromaticity coordinates corresponding to every pixel in the tristimulus image im (HxWx3). [c1. The O and A vectors are normalized and the normal vector is formed from O x A. trnorm Normalize a homogeneous transform tn = trnorm(T) is a normalized homogeneous transformation matrix in which the rotation submatrix is guaranteed to be a proper orthogonal matrix. If tri is RGB then cc is rg. Multiple tristimulus values can be given as rows of tri (N × 3) in which case the chromaticity coordinates are the corresponding rows of cc (N × 2). if tri is XYZ then cc is xy. Notes • somewhat unusually this function performs a function and its inverse. FUNCTIONS AND CLASSES transform sequence and returns an N × 3 matrix where each row is the translational component of the corresponding transform in the sequence. See also ctraj tristim2cc Tristimulus to chromaticity coordinates cc = tristim2cc(tri) is the chromaticity coordinate (1 × 2) corresponding to the tristimulus tri (1 × 3).

troty.CHAPTER 2. FUNCTIONS AND CLASSES Notes • Used to prevent ﬁnite word length arithmetic causing transforms to become ‘unnormalized’. Notes • translational component is zero See also rotx. See also oa2tr trotx Rotation about X axis T = trotx(theta) is a homogeneous transformation representing a rotation of theta about the x-axis. trotz troty Rotation about Y axis T = troty(theta) is a homogeneous transformation representing a rotation of theta about the y-axis. Notes • translational component is zero Machine Vision Toolbox for MATLAB 182 R Copyright c Peter Corke 2011 .

Notes • translational component is zero See also rotz. f ‘label’. Options ‘rpy’ ‘euler’ ‘angvec’ ‘radian’ ‘fmt’. l display with rotation in roll/pitch/yaw angles (default) display with rotation in ZYX Euler angles display with rotation in angle/vector format display angle in radians (default is degrees) use format string f for all numbers. options) displays the homogoneous transform in a compact single-line format. trprint T is the command line form of above. trotx. (default %g) display the text before the transform Machine Vision Toolbox for MATLAB 183 R Copyright c Peter Corke 2011 . and displays in RPY format.CHAPTER 2. FUNCTIONS AND CLASSES See also roty. troty trprint Compact display of homogeneous transformation trprint(T. trotx. If T is a homogeneous transform sequence then print each element is printed on a separate line. trotz trotz Rotation about Z axis T = trotz(theta) is a homogeneous transformation representing a rotation of theta about the z-axis.

(x-x0)p . FUNCTIONS AND CLASSES See also tr2eul.y0) is the centroid.y). That is. Note • fails for the case where norm(v) is zero.CHAPTER 2. q) is the PQ’th central moment of the image im.(y-y0)q where (x0. tr2rpy. mpq. Notes • The central moments are invariant to translation. See also upq poly. the sum of I(x. upq Central image moments m = upq(im. npq Machine Vision Toolbox for MATLAB 184 R Copyright c Peter Corke 2011 . p. tr2angvec unit Unitize a vector vn = unit(v) is a unit vector parallel to v.

CHAPTER 2. FUNCTIONS AND CLASSES

upq poly

Central polygon moments

m = UPQ POLY(v, p, q) is the PQ’th central moment of the polygon with vertices described by the columns of v.

Notes

• The points must be sorted such that they follow the perimeter in sequence (counterclockwise). • If the points are clockwise the moments will all be negated, so centroids will be still be correct. • If the ﬁrst and last point in the list are the same, they are considered as a single vertex. • The central moments are invariant to translation.

See also

upq, mpq poly, npq poly

useﬁg

a named ﬁgure or create a new ﬁgure

useﬁg(’Foo’) make ﬁgure ‘Foo’ the current ﬁgure, if it doesn’t exist create it. h = useﬁg(’Foo’) as above, but returns the ﬁgure handle

vex

Convert skew-symmetric matrix to vector

v = vex(s) is the vector which has the skew-symmetric matrix s.

Machine Vision Toolbox for MATLAB 185

R

Copyright c Peter Corke 2011

CHAPTER 2. FUNCTIONS AND CLASSES

Notes

• No checking is done to ensure that the matrix is skew-symmetric. • The function takes the mean of the two elements that correspond to each unique element of the matrix, ie. vx = 0.5*(s(3,2)-s(2,3))

See also

skew

xaxis

X-axis scaling

xaxis(max) xaxis([min max]) xaxis(min, max) xaxis restore automatic scaling for this axis

xycolorspace

Display spectral locus

xycolorspace() display a fully colored spectral locus in terms of CIE x and y coordinates. xycolorspace(p) as above but plot the points whose xy-chromaticity is given by the columns of p. [im,ax,ay] = xycolorspace() as above returns the spectral locus as an image im, with corresponding x- and y-axis coordinates ax and ay respectively.

Notes

• The colors shown within the locus only approximate the true colors, due to the gamut of the display device.

Machine Vision Toolbox for MATLAB 186

R

Copyright c Peter Corke 2011

CHAPTER 2. FUNCTIONS AND CLASSES

See also

rg addticks

yaxis

Y-axis scaling

yayis(max) yayis(min, max) YAXIS restore automatic scaling for this axis

zcross

Zero-crossing detector

iz = zcross(im) is a binary image with pixels set where the corresponding pixels in the signed image im have a zero crossing, a positive pixel adjacent to a negative pixel.

Notes

• Can be used in association with a Lapalacian of Gaussian image to determine edges.

See also

ilog

Machine Vision Toolbox for MATLAB 187

R

Copyright c Peter Corke 2011

The result m is a scalar in the interval -1 to 1 that indicates similarity. ssd. The result m is a scalar that indicates image similarity. isimilarity zsad Sum of absolute differences m = zsad(i1. i2) is the zero-mean sum of absolute differences between the two equally sized image patches i1 and i2. See also sad.CHAPTER 2. See also ncc. A value of 1 indicates identical pixel patterns. FUNCTIONS AND CLASSES zncc Normalized cross correlation m = zncc(i1. ssd. a value of 0 indicates identical pixel patterns and is increasingly positive as image dissimilarity increases. sad. Notes • The zsad similarity measure is invariant to changes in image brightness offset. isimilarity Machine Vision Toolbox for MATLAB 188 R Copyright c Peter Corke 2011 . i2) is the zero-mean normalized cross-correlation between the two equally sized image patches i1 and i2. ncc. Notes • The zncc similarity measure is invariant to afﬁne changes in image intensity (brightness offset and scale).

ncc. See also sdd. a value of 0 indicates identical pixel patterns and is increasingly positive as image dissimilarity increases. i2) is the zero-mean sum of squared differences between the two equally sized image patches i1 and i2. FUNCTIONS AND CLASSES zssd Sum of squared differences m = zssd(i1.CHAPTER 2. sad. Notes • The zssd similarity measure is invariant to changes in image brightness offset. isimilarity Machine Vision Toolbox for MATLAB 189 R Copyright c Peter Corke 2011 . The result m is a scalar that indicates image similarity.