Download textbook Computer Vision Methods For Fast Image Classi%Ef%Ac%81Cation And Retrieval Rafal Scherer ebook all chapter pdf

Computer Vision Methods for Fast
Image Classi■cation and Retrieval

Rafa■ Scherer
Visit to download the full and correct content document:
https://textbookfull.com/product/computer-vision-methods-for-fast-image-classi%ef%a
c%81cation-and-retrieval-rafal-scherer/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Feature extraction and image processing for computer

vision Fourth Edition Aguado
https://textbookfull.com/product/feature-extraction-and-image-
processing-for-computer-vision-fourth-edition-aguado/
Feature Extraction and Image Processing for Computer

Vision 4th Edition Mark Nixon
https://textbookfull.com/product/feature-extraction-and-image-
processing-for-computer-vision-4th-edition-mark-nixon/
Image Processing and Computer Vision in iOS Oge Marques
https://textbookfull.com/product/image-processing-and-computer-
vision-in-ios-oge-marques/
Nonlinear Eigenproblems in Image Processing and

Computer Vision Guy Gilboa
https://textbookfull.com/product/nonlinear-eigenproblems-in-
image-processing-and-computer-vision-guy-gilboa/
Computer Vision Pattern Recognition Image Processing
and Graphics Renu Rameshan
https://textbookfull.com/product/computer-vision-pattern-
recognition-image-processing-and-graphics-renu-rameshan/
Image Processing Computer Vision and Pattern

Recognition Hamid R. Arabnia
https://textbookfull.com/product/image-processing-computer-
vision-and-pattern-recognition-hamid-r-arabnia/
Computer Vision and Image Analysis Digital Image

Processing and Analysis 4th Edition Scott E Umbaugh
https://textbookfull.com/product/computer-vision-and-image-
analysis-digital-image-processing-and-analysis-4th-edition-scott-
e-umbaugh/
Numerical Algorithms Methods for Computer Vision

Machine Learning and Graphics 1st Edition Justin
Solomon
https://textbookfull.com/product/numerical-algorithms-methods-
for-computer-vision-machine-learning-and-graphics-1st-edition-
justin-solomon/
Energy Minimization Methods in Computer Vision and

Pattern Recognition Marcello Pelillo
https://textbookfull.com/product/energy-minimization-methods-in-
computer-vision-and-pattern-recognition-marcello-pelillo/
Studies in Computational Intelligence 821
Rafał Scherer
Computer
Vision Methods
for Fast Image
Classification
and Retrieval
Studies in Computational Intelligence
Volume 821
Series editor
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
e-mail: kacprzyk@ibspan.waw.pl
The series “Studies in Computational Intelligence” (SCI) publishes new develop-
ments and advances in the various areas of computational intelligence—quickly and
with a high quality. The intent is to cover the theory, applications, and design
methods of computational intelligence, as embedded in the fields of engineering,
computer science, physics and life sciences, as well as the methodologies behind
them. The series contains monographs, lecture notes and edited volumes in
computational intelligence spanning the areas of neural networks, connectionist
systems, genetic algorithms, evolutionary computation, artificial intelligence,
cellular automata, self-organizing systems, soft computing, fuzzy systems, and
hybrid intelligent systems. Of particular value to both the contributors and the
readership are the short publication timeframe and the world-wide distribution,
which enable both wide and rapid dissemination of research output.
The books of this series are submitted to indexing to Web of Science,
EI-Compendex, DBLP, SCOPUS, Google Scholar and Springerlink.
More information about this series at http://www.springer.com/series/7092

Rafał Scherer
Computer Vision Methods

for Fast Image Classification
and Retrieval
123
Rafał Scherer
Institute of Computational Intelligence
Częstochowa University of Technology
Częstochowa, Poland
ISSN 1860-949X ISSN 1860-9503 (electronic)

Studies in Computational Intelligence
ISBN 978-3-030-12194-5 (hardcover) ISBN 978-3-030-12195-2 (eBook)
ISBN 978-3-030-12197-6 (softcover)
https://doi.org/10.1007/978-3-030-12195-2
Library of Congress Control Number: 2018968376
© Springer Nature Switzerland AG 2020

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Computer vision and image retrieval and classification are a vital set of methods
used in various engineering, scientific and business applications. In order to
describe an image, visual features must be detected and described. Usually, the
description is in the form of vectors. The book presents methods for accelerating
image retrieval and classification in large datasets. Some of the methods (Chap. 5)
are designed to work directly in relational database management systems.
The book is the result of collaboration with colleagues from the Institute of
Computational Intelligence at the Częstochowa University of Technology. I would
like to thank my former Ph.D. students Dr. Rafał Grycuk and Dr. Patryk Najgebauer
for their cooperation.
I would like to express my sincere thanks to my friend Prof. Marcin
Korytkowski for his invaluable help in research and to Prof. Leszek Rutkowski,
who introduced me to scientific work and supported me in a friendly manner. I am
also grateful to the Institute of Computational Intelligence at the Częstochowa
University of Technology for providing a scholarly environment for both teaching
and research.
Finally, I am truly grateful to my wife Magda, my children Karolina and
Katarzyna for their love and patience and to my mother for raising me in the way
that she did.
Częstochowa, Poland Rafał Scherer

November 2018
v
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Feature Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Local Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Scale-Invariant Feature Transform (SIFT) . . . . . . . . . . . . . 7
2.1.2 Speed-Up Robust Features (SURF) . . . . . . . . . . . . . . . . . . 9
2.2 Edge Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Canny Edge Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Blob Detection and Blob Extraction . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Clustering Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.1 K-means Clustering Algorithm . . . . . . . . . . . . . . . . . . . . . 20
2.4.2 Mean Shift Clustering Algorithm . . . . . . . . . . . . . . . . . . . 20
2.5 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.6 Global Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6.1 Colour and Edge Directivity CEDD Descriptor . . . . . . . . . 27
2.7 Summary and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3 Image Indexing Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1 Image Classification by Fuzzy Rules . . . . . . . . . . . . . . . . . . . . . . 34
3.1.1 Boosting-Generated Simple Fuzzy Classifiers . . . . . . . . . . 36
3.1.2 Classification of a Query Image . . . . . . . . . . . . . . . . . . . . 40
3.1.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.1.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Salient Object Detector and Descriptor by Edge Crawler . . . . . . . 44
3.2.1 System for Content-Based Image Retrieval . . . . . . . . . . . . 47
3.2.2 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
vii
viii Contents
3.3 Fast Two-Level Image Indexing . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.3.1 Hash Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.2 Structure of the Proposed Descriptor Index . . . . . . . . . . . . 58
3.3.3 Index Search Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.4 Image Colour Descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.4.1 Method Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.4.2 Color Descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4.3 Colour Relationship Sub-descriptor . . . . . . . . . . . . . . . . . . 67
3.4.4 Descriptor Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.5 Fast Dictionary Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.5.1 Description of the Problem . . . . . . . . . . . . . . . . . . . . . . . . 72
3.5.3 Comparison Between Descriptors and Dictionary . . . . . . . . 73
3.5.4 Matching Sets of Keypoints . . . . . . . . . . . . . . . . . . . . . . . 74
3.5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4 Novel Methods for Image Description . . . . . . . . . . . . . . . . . . . . . . . . 83
4.1 Algorithm for Discontinuous Edge Description . . . . . . . . . . . . . . . 83
4.1.1 Proposed Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.1.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2 Interest Point Localization Based on the Gestalt Laws . . . . . . . . . 95
4.2.1 Problem Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5 Image Retrieval and Classification in Relational Databases . . . . . . . 107
5.1 Bag of Features Image Classification in Relational Databases . . . . 108
5.1.1 System Architecture and Relational Database Structure . . . 109
5.1.2 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.1.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Contents ix
5.2 Bag of Features Image Retrieval in Relational Databases . . . . . . . 114

5.2.1 Description of the Proposed System . . . . . . . . . . . . . . . . . 114
5.2.2 Numerical Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.2.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.3 Database Indexing System Based on Boosting and Fuzzy
Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.3.1 Building Visual Index . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.3.2 Proposed Database Framework . . . . . . . . . . . . . . . . . . . . . 120
5.3.3 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.4 Database Retrieval System Based on the CEDD Descriptor . . . . . . 126
5.4.1 Simulation Environment . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.4.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6 Concluding Remarks and Perspectives in Computer Vision . . . . . . . 137
Chapter 1
Introduction
In recent times, one can observe the increasing development of multimedia tech-
nologies and their rising dominance in life and business. Society is becoming more
eager to use new solutions as they facilitate life, primarily by simplifying contact
and accelerating the exchange of experience with others, what was not encountered
on such a large scale many years ago.
Computer vision solutions are being developed increasingly to oversee produc-
tion processes in order to ensure their correct operation. Until now, most of them
could only be properly supervised by humans. Control requires focusing and con-
sists in constantly performing identical activities. Work monotony lowers human
concentration, which is more likely to make a mistake or overlook important facts.
Healthcare, and in particular medical diagnostics, is one of the areas that provide
a relatively broad spectrum of possible applications for computer vision solutions. In
the past, most methods focused on processing and delivery of results in the most read-
able form to the doctor’s diagnosis for analysis. These include medical imaging, such
as computed tomography, magnetic resonance and ultrasonography, which transform
signals from the device into a diagnostic readable image. Now, the diagnosis can be
automatised thanks to image classification.
The most popular way to search vast collections of images and video which are
generated every day in a tremendous amount is realized by keywords and meta
tags or just by browsing them. Emergence of content-based image retrieval (CBIR)
in the 1990s enabled automatic retrieval of images to a certain extent. Various
CBIR tasks include searching for images similar to the query image or retrieving
images of a certain class [11, 20, 21, 28, 29, 31, 41, 50, 51, 53] and classification
[2, 6, 10, 18, 19, 22, 30, 44, 52] of the query image. Such content-based image
matching remains a challenging problem of computer science. Image matching con-
sists of two relatively difficult tasks: identifying objects in images and fast searching
through large collections of identified objects. Identifying objects on images is still
© Springer Nature Switzerland AG 2020 1

R. Scherer, Computer Vision Methods for Fast Image Classification
and Retrieval, Studies in Computational Intelligence 821,
https://doi.org/10.1007/978-3-030-12195-2_1
2 1 Introduction
a challenge as the same objects and scenes can be viewed under different imaging
conditions. There are many previous works dedicated to the problem formulated in
this way. Some of them are based on color representation [15, 25, 39], textures [9, 12,
17, 46], shape [16, 23, 49] or edge detectors [37, 38, 56]. Local invariant features
have gained a wide popularity [32–34, 36, 45]. The most popular local keypoint
detectors and descriptors are SURF [4], SIFT [32] or ORB [42].
In content-based image retrieval and classification, we can distinguish two
approaches. The first one gradually generalises information from an image. To this
group, we can include methods based on machine learning such as convolutional
neural networks, e.g. [8, 26] or older methods based on histograms [40, 48]. These
methods try to reduce the amount of visual feature data to describe the entire image at
the highest possible level. Neural networks can be trained to recognise and classify
particular elements of an image, but they lose some information that is crucial to
determine if the content between images is identical.
To check similarity between images we can use methods from the second popular
group that is based on local interest points (keypoints), or other features that describe
the local content of an image. Such methods do not generalise the content of an image
and do not try to classify it. They usually generate significant amount of data, but
they can find similar fragments of content between images. Thanks to this, this
group of methods found multiple applications in video tracking and processing, for
example, to correct content transition between frames during the camera move [14,
55]. Another popular application is a three-dimensional object reconstruction from
a set of images. Some popular methods include SIFT, SURF, HOG, ORB, BRIEF,
FREAK, with many modifications [3, 5, 24, 27, 43, 47, 54].
In the case of the first group of methods, work with a larger set of images is
easier, because the result features are simple and in most cases can be easily stored
and searched. But in the case of the second group, the main problem is a large and
variable amount of data per image. To speed up the search process, we can use
methods that can learn keypoint structure or descriptors [7, 13].
Developing content-based image comparison methods that simulate human visual
perception is a very hard and complicated process. Image recognition is natural
and very simple for humans but when we try to mimic the process we face many
problems as it is very complicated, uses multiple hidden techniques developed during
the evolution and we only have a rough sense of how the brain works. Most of them,
e.g. human imagination, are currently unavailable for computer systems. Also huge
knowledge, which humans acquire through the entire life is hard to store for machine
learning systems and we excel in visual identification. Thus, image comparison
algorithms try to extract and simplify this large amount of data from images to form
a structurized description that is easy to compare for computers, such as human text
writing [1, 35]. But image description is extracted only from the image pixel spatial
distribution and is not supported by human imagination or knowledge. That caused
that image description in most cases is not fully satisfactory for human users.
The book presents some methods for accelerating image retrieval and classifica-
tion in large collections of images using so-called hand-crafted features.
References 3
References
1. Aghdam, M.H., Heidari, S.: Feature selection using particle swarm optimization in text cate-
gorization. J. Artif. Intell. Soft Comput. Res. 5(4), 231–238 (2015)
2. Akusok, A., Miche, Y., Karhunen, J., Bjork, K.M., Nian, R., Lendasse, A.: Arbitrary category
classification of websites based on image content. Comput. Intell. Mag. IEEE 10(2), 30–41
(2015). https://doi.org/10.1109/MCI.2015.2405317
3. Alahi, A., Ortiz, R., Vandergheynst, P.: Freak: Fast retina keypoint. In: 2012 IEEE Conference
on Computer Vision and Pattern Recognition, pp. 510–517. IEEE (2012)
4. Bay, H., Ess, A., Tuytelaars, T., Van Gool, L.: Speeded-up robust features (surf). Comput. Vis.
Image Underst. 110(3), 346–359 (2008)
5. Bay, H., Tuytelaars, T., Van Gool, L.: Surf: Speeded up robust features. In: Computer vision–
ECCV 2006, pp. 404–417. Springer (2006)
6. Bazarganigilani, M.: Optimized image feature selection using pairwise classifiers. J. Artif.
Intell. Soft Comput. Res. 1(2), 147–153 (2011)
7. Calonder, M., Lepetit, V., Fua, P.: Keypoint signatures for fast learning and recognition. In:
European Conference on Computer Vision, pp. 58–71. Springer (2008)
8. Chang, O., Constante, P., Gordon, A., Singana, M.: A novel deep neural network that uses
space-time features for tracking and recognizing a moving object. J. Artif. Intell. Soft Comput.
Res. 7(2), 125–136 (2017)
9. Chang, T., Kuo, C.C.: Texture analysis and classification with tree-structured wavelet transform.
Image Process. IEEE Trans. 2(4), 429–441 (1993). https://doi.org/10.1109/83.242353
10. Chang, Y., Wang, Y., Chen, C., Ricanek, K.: Improved image-based automatic gender classi-
fication by feature selection. J. Artif. Intell. Soft Comput. Res. 1(3), 241–253 (2011)
11. Daniel Carlos Guimaraes Pedronette, J.A., da S. Torres, R.: A scalable re-ranking method for
content-based image retrieval. Inf. Sci. 265(0), 91–104 (2014). https://doi.org/10.1016/j.ins.
2013.12.030
12. Francos, J., Meiri, A., Porat, B.: A unified texture model based on a 2-d wold-like decom-
position. Signal Process. IEEE Trans. 41(8), 2665–2678 (1993). https://doi.org/10.1109/78.
229897
13. Grabner, M., Grabner, H., Bischof, H.: Learning features for tracking. In: IEEE Conference on
Computer Vision and Pattern Recognition, 2007 CVPR 2007, pp. 1–8. IEEE (2007)
14. Hare, S., Saffari, A., Torr, P.H.: Efficient online structured output learning for keypoint-based
object tracking. In: 2012 IEEE Conference on Computer Vision and Pattern Recognition
(CVPR), pp. 1894–1901. IEEE (2012)
15. Huang, J., Kumar, S., Mitra, M., Zhu, W.J., Zabih, R.: Image indexing using color correlograms.
In: Proceedings of 1997 IEEE Computer Society Conference on Computer Vision and Pattern
Recognition, 1997, pp. 762–768 (1997). https://doi.org/10.1109/CVPR.1997.609412
16. Jagadish, H.V.: A retrieval technique for similar shapes. SIGMOD Rec. 20(2), 208–217 (1991)
17. Jain, A.K., Farrokhnia, F.: Unsupervised texture segmentation using gabor filters. Pattern
Recogn. 24(12), 1167–1186 (1991)
18. Jégou, H., Douze, M., Schmid, C., Pérez, P.: Aggregating local descriptors into a compact
image representation. In: 2010 IEEE Conference on Computer Vision and Pattern Recognition
(CVPR), pp. 3304–3311. IEEE (2010)
19. Jégou, H., Perronnin, F., Douze, M., Sanchez, J., Perez, P., Schmid, C.: Aggregating local
image descriptors into compact codes. Pattern Anal. Mach. Intell. IEEE Trans. 34(9), 1704–
1716 (2012)
20. Kanimozhi, T., Latha, K.: An integrated approach to region based image retrieval using firefly
algorithm and support vector machine. Neurocomputing 151, Part 3(0), 1099–1111 (2015)
21. Karakasis, E., Amanatiadis, A., Gasteratos, A., Chatzichristofis, S.: Image moment invariants
as local features for content based image retrieval using the bag-of-visual-words model. Pattern
Recogn. Lett. 55, 22–27 (2015)
4 1 Introduction
22. Karimi, B., Krzyzak, A.: A novel approach for automatic detection and classification of sus-
picious lesions in breast ultrasound images. J. Artif. Intell. Soft Comput. Res. 3(4), 265–276
(2013)
23. Kauppinen, H., Seppanen, T., Pietikainen, M.: An experimental comparison of autoregressive
and fourier-based descriptors in 2d shape classification. Pattern Anal. Mach. Intell. IEEE Trans.
17(2), 201–207 (1995). https://doi.org/10.1109/34.368168
24. Ke, Y., Sukthankar, R.: Pca-sift: A more distinctive representation for local image descriptors.
In: Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and
Pattern Recognition, 2004. CVPR 2004. , vol. 2, pp. II–II. IEEE (2004)
25. Kiranyaz, S., Birinci, M., Gabbouj, M.: Perceptual color descriptor based on spatial distribution:
a top-down approach. Image Vision Comput. 28(8), 1309–1326 (2010)
26. Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional
neural networks. Adv. Neural Inf. Process. Syst., 1097–1105 (2012)
27. Leutenegger, S., Chli, M., Siegwart, R.Y.: Brisk: Binary robust invariant scalable keypoints.
In: 2011 IEEE International Conference on Computer Vision (ICCV), pp. 2548–2555. IEEE
(2011)
28. Lin, C.H., Chen, H.Y., Wu, Y.S.: Study of image retrieval and classification based on adaptive
features using genetic algorithm feature selection. Expert Syst. Appl. 41(15), 6611–6621 (2014)
29. Liu, G.H., Yang, J.Y.: Content-based image retrieval using color difference histogram. Pattern
Recogn. 46(1), 188–198 (2013)
30. Liu, L., Shao, L., Li, X.: Evolutionary compact embedding for large-scale image classification.
Inf. Sci. 316, 567–581 (2015)
31. Liu, S., Bai, X.: Discriminative features for image classification and retrieval. Pattern Recogn.
Lett. 33(6), 744–751 (2012)
32. Lowe, D.G.: Distinctive image features from scale-invariant keypoints. Int. J. Comput. Vis.
60(2), 91–110 (2004)
33. Matas, J., Chum, O., Urban, M., Pajdla, T.: Robust wide-baseline stereo from maximally
stable extremal regions. Image Vis. Comput. 22(10), 761–767 (2004). British Machine Vision
Computing 2002
34. Mikolajczyk, K., Schmid, C.: Scale and affine invariant interest point detectors. Int. J. Comput.
Vis. 60(1), 63–86 (2004)
35. Murata, M., Ito, S., Tokuhisa, M., Ma, Q.: Order estimation of japanese paragraphs by super-
vised machine learning and various textual features. J. Artif. Intell. Soft Comput. Res. 5(4),
247–255 (2015)
36. Nister, D., Stewenius, H.: Scalable recognition with a vocabulary tree. In: Proceedings of
the 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition -
Volume 2, CVPR ’06, pp. 2161–2168. IEEE Computer Society, Washington, DC, USA (2006)
37. Ogiela, M.R., Tadeusiewicz, R.: Syntactic reasoning and pattern recognition for analysis of
coronary artery images. Artif. Intell. Med. 26(1), 145–159 (2002)
38. Ogiela, M.R., Tadeusiewicz, R.: Nonlinear processing and semantic content analysis in medical
imaging-a cognitive approach. Instrum. Measurement IEEE Trans. 54(6), 2149–2155 (2005)
39. Pass, G., Zabih, R.: Histogram refinement for content-based image retrieval. In: Proceedings
3rd IEEE Workshop on Applications of Computer Vision, 1996. WACV 1996, pp. 96–102
(1996). https://doi.org/10.1109/ACV.1996.572008
40. Pass, G., Zabih, R., Miller, J.: Comparing images using color coherence vectors. In: Proceedings
of the fourth ACM international conference on Multimedia, pp. 65–73. ACM (1997)
41. Rashedi, E., Nezamabadi-pour, H., Saryazdi, S.: A simultaneous feature adaptation and feature
selection method for content-based image retrieval systems. Knowl. Based Syst. 39, 85–94
(2013)
42. Rublee, E., Rabaud, V., Konolige, K., Bradski, G.: Orb: An efficient alternative to sift or surf.
In: 2011 IEEE International Conference on Computer Vision (ICCV), pp. 2564–2571 (2011).
https://doi.org/10.1109/ICCV.2011.6126544
43. Rublee, E., Rabaud, V., Konolige, K., Bradski, G.: Orb: an efficient alternative to sift or surf.
In: 2011 IEEE International Conference on Computer Vision (ICCV), pp. 2564–2571. IEEE
(2011)
References 5
44. Shrivastava, N., Tyagi, V.: Content based image retrieval based on relative locations of multiple
regions of interest using selective regions matching. Inf. Sci. 259, 212–224 (2014). https://doi.
org/10.1016/j.ins.2013.08.043
45. Sivic, J., Zisserman, A.: Video google: a text retrieval approach to object matching in videos.
In: Proceedings of Ninth IEEE International Conference on Computer Vision, 2003, pp. 1470–
1477 vol. 2 (2003)
46. Śmietański, J., Tadeusiewicz, R., Łuczyńska, E.: Texture analysis in perfusion images of
prostate cancera case study. Int. J. Appl. Math. Comput. Sci. 20(1), 149–156 (2010)
47. Sünderhauf, N., Protzel, P.: Brief-gist-closing the loop by simple means. In: 2011 IEEE/RSJ
International Conference on Intelligent Robots and Systems (IROS), pp. 1234–1241. IEEE
(2011)
48. Tsai, G.: Histogram of oriented gradients. Univ. Mich. 1(1), 1–17 (2010)
49. Veltkamp, R.C., Hagedoorn, M.: State of the art in shape matching. In: Lew, M.S. (ed.) Prin-
ciples of Visual Information Retrieval, pp. 87–119. Springer, London, UK, UK (2001)
50. Wang, X.Y., Yang, H.Y., Li, Y.W., Li, W.Y., Chen, J.W.: A new svm-based active feedback
scheme for image retrieval. Eng. Appl. Artif. Intell. 37, 43–53 (2015)
51. Wu, J., Shen, H., Li, Y.D., Xiao, Z.B., Lu, M.Y., Wang, C.L.: Learning a hybrid similarity
measure for image retrieval. Pattern Recogn. 46(11), 2927–2939 (2013)
52. Yang, J., Yu, K., Gong, Y., Huang, T.: Linear spatial pyramid matching using sparse coding
for image classification. In: IEEE Conference on Computer Vision and Pattern Recognition,
2009. CVPR 2009, pp. 1794–1801 (2009). https://doi.org/10.1109/CVPR.2009.5206757
53. Yu, J., Qin, Z., Wan, T., Zhang, X.: Feature integration analysis of bag-of-features model
for image retrieval. Neurocomputing 120(0), 355–364 (2013). Image Feature Detection and
Description
54. Žemgulys, J., Raudonis, V., Maskeliūnas, R., Damaševičius, R.: Recognition of basketball
referee signals from videos using histogram of oriented gradients (hog) and support vector
machine (svm). Procedi. Comput. Sci. 130, 953–960 (2018)
55. Zhao, W.L., Ngo, C.W.: Scale-rotation invariant pattern entropy for keypoint-based near-
duplicate detection. IEEE Trans. Image Process. 18(2), 412–423 (2009)
56. Zitnick, C., Dollar, P.: Edge boxes: Locating object proposals from edges. In: Fleet, D., Pajdla,
T., Schiele, B., Tuytelaars, T., (eds.) Computer Vision ECCV 2014, Lecture Notes in Computer
Science, vol. 8693, pp. 391–405. Springer International Publishing (2014)
Chapter 2
Feature Detection
Computer vision relies on image features describing points, edges, objects or colour.
The book concerns solely so-called hand-made features contrary to learned features
which exist in deep learning methods. Image features can be generally divided into
global and local methods.
Global methods extract features from the entire image without dividing into more
and less significant areas. To this group, we can include histogram-based algorithms
such as histogram of oriented gradients (HOG) or colour coherence vector (CCV)
[12, 47]. In most cases, they generate a constant amount of description data which is
easier to compare and store, on the other hand, image comparison by histogram-based
algorithms gives only a vague similarity for a user.
Local feature-based methods try at first to find significant characteristic areas of
an image based on Laplacian of Gaussian (LoG) or Difference of Gaussian (DoG)
algorithms [25, 64]. And then they generate a description of their neighbourhood.
These methods are more accurate, on the other hand, can generate a lot of description
data and that amount varies per image. Local feature methods based on keypoints
are efficient in similarity detection between images but less in content recognition.
Commonly used methods of this kind are SIFT, SURF, ORB, BRIEF, FAST [4, 6,
46, 48, 49].
2.1 Local Features
2.1.1 Scale-Invariant Feature Transform (SIFT)
SIFT (Scale-Invariant Feature Transform) is an algorithm used to detect and describe

local features of an image. It was presented for the first time in [37] and is now
© Springer Nature Switzerland AG 2020 7

R. Scherer, Computer Vision Methods for Fast Image Classification
and Retrieval, Studies in Computational Intelligence 821,
https://doi.org/10.1007/978-3-030-12195-2_2
8 2 Feature Detection
patented by the University of British Columbia. For each keypoint, which describes
the local image feature, we generate a feature vector, that can be used for further
processing. The algorithm is immune to changing scale, rotation and light change.
SIFT consists of the four main steps [36]
1. Scale-space extreme detection—Extraction of potential keypoints by scanning
the entire image,
Constructing scale-space,
Laplacian approximation by Gaussian blur,
2. Keypoint localization—Selection of stable keypoints (resistant to change of scale
and rotation),
Removing not important keypoints (noise).
3. Orientation assignment—Finding keypoint orientation resistant to the image
transformation,
4. Keypoint descriptor—Generating vectors describing keypoints.
During the process of creating scale-space, the image is rescaled (creating octaves) in
order to detect the most important and resistant features. After this step, a scale-space
pyramid is obtained. This pyramid consists of octaves, sorted from the largest to the
smallest octave.
In the next stage, Gaussian blur is applied. This step is performed by the following
Gaussian operator [37]
L(x, y, σ) = G(x, y, σ) ∗ I (x, y), (2.1)
where L is the output image, G represents the Gaussian operator, I is the input image.
In the next step, the Laplacian is calculated, in order to detect edges. It should be
done by calculation of the second derivative, but this operation is computationally
expensive. In order to overcome this nuisance, the Difference of Gaussians (DOG)
is performed. The next stage of the SIFT algorithm is the localisation of keypoints.
This step consists of two important steps
• localisation of local extrema on DoG images,
• setting the extrema position.
The localisation of local extrema is based on comparing the pixels with their neigh-
bours. On a discrete image, the brightest pixel not always has the same position as
local extrema. Thus, this issue was solved by using Taylor’s theorem
∂ DT 1 ∂2 D
D(x) = D + x + xT x. (2.2)
∂x 2 ∂x 2
Each keypoint contains two features: strength and orientation in which keypoint is
directed. It is calculated by gradients of its neighbours. Each resultant SIFT keypoint
descriptor consists of two vectors. The first one contains the point position (x, y),
2.1 Local Features 9
scale (detected scale), response (response of the detected feature, strength), orienta-
tion (orientation measured anti-clockwise from +ve x-axis), the Laplacian (sign of
the Laplacian for fast matching purposes). The second one contains the descriptor
of 128 length.
2.1.2 Speed-Up Robust Features (SURF)
SURF (Speeded-Up Robust Features) is a method which allows to detect and describe
local features of an image. SURF is an improved version of SIFT in terms of speed
(see Sect. 2.1.1) [37]. It was presented for the first time in [4], nowadays is widely
used in various systems e.g. image recognition, object description [24], segmentation
[23], image analysis [24] and image retrieval [22, 57], object tracking [21], and many
others. SURF is similar algorithm to SIFT. The Integral Images are used instead of
DOG (Difference of Gaussian), which allows it to work much faster than SIFT. An
important advantage of SURF is that it generates less data than SIFT (SURF has a
shorter descriptor of 64 length), which speeds up further processing. The algorithm
has also a parallel version [53, 56] thus, it generates the results much faster. SURF
generates image keypoints (interesting points) which allows to extract and match
local features, e.g. in pairs of images. For each keypoint, which indicates a local
image feature, a feature vector is generated. It describes the keypoint surrounding
and allows to determine its orientation. These vectors are often used for further
processing in many computer vision methods. SURF consists of four main steps:
• Computing Integral Images,
• Fast Hessian Detector,
– The Hessian,
– Constructing the Scale-Space,
– Accurate Interest Point Localization,
• Interest Point Descriptor,
– Orientation Assignment,
– Descriptor Components,
• Generating vectors describing the keypoint.
In the first step, the integral images are calculated. It allows to increase the efficiency.
This method is very simple and it is used for calculation of pixels sum in the given
rectangle area. This process can be described by the following formula [15]

i<x
j<y
I (x, y) = I (x, y), (2.3)
i=0 j=0
Fig. 2.1 Calculating area

using integral images
where I is a processed image and I (x, y) is a sum of pixels in given area. The
usage of integral images in order to calculating area is reduced to four operation. Let
us consider a rectangle described by vertices A, B, C, D. The example is presented
in Fig. 2.1. The sum of pixel intensities is calculated by formula [15]

= A + D − (C + B). (2.4)
In the next step, the Hessian matrix determinant is calculated. The Hessian matrix is
presented below [15]
∂2 f ∂2 f
∂x 2 ∂x∂ y
H ( f (x, y)) = ∂2 f ∂2 f . (2.5)
∂x∂ y ∂ y 2
The determinant of this matrix can be calculated by [15]

∂2 f ∂2 f ∂2 f
det (H ) = − . (2.6)
∂x 2 ∂ y 2 ∂x∂ y
The calculation of the local maximum by the Hessian matrix determinant depends
from the sign of this determinant. If its value is greater or equal 0, this area is
determined as the local maximum. In the next step, the scale-space is constructed.
This step is used in order to make the keypoint immune to changing scale and
rotation. In the interest point localization stage the min H essian parameter is needed.
To determine the threshold value. The localization is calculated by comparing the
Hessian determinant with its neighbours.
The process of creating the keypoint descriptors is performed by using Haar
wavelets (see Fig. 2.2) which describe the keypoints gradients. In order to compute
the descriptor orientation, the algorithm searches the largest Haar wavelets sum in
π/3 (60) window and step +−15 (see Fig. 2.3).
2.1 Local Features 11
Fig. 2.2 Haar wavelets. These filters calculate responses for x (left) and y (right) directions
Fig. 2.3 Descriptor orientation assignment. The window of π/3 size moves around the origin and
determines the sum of largest wavelets sum, which allows obtaining the longest vector [15]
Fig. 2.4 Visual

representation of the SURF
descriptor. The descriptor
contains 16 subregions and
each of them consists of
5 × 5 calculated Haar
wavelets [15]
The descriptor consists of 4 × 4 wavelet group matrix. Each of them is composed of

the calculated Haar wavelets and is divided into 5 × 5 elements (Fig. 2.4).
The output SURF keypoint consists of two vectors. The first one contains: point
position (x, y), scale (detected scale), response (response of the detected feature,
strength), orientation (orientation measured anti-clockwise from +ve x-axis), the
Laplacian (the sign of the Laplacian for fast matching purposes). The second one
describes the intensity distribution of the pixels within the neighbourhood of the
interest point (64 values). In order to generate keypoints, SURF requires one input
parameter—min H essian. During the experiments, this variable was usually set to
Fig. 2.5 The SURF algorithm example with keypoint detection and matching
Fig. 2.6 Another example of keypoint matching by the SURF algorithm
400. This value was obtained empirically, and it was suitable in many experiments.
The method is immune to change of scale and rotation, which allows for matching
corresponding keypoints in similar images [27, 43]. Figure 2.5 shows an example of
the SURF method with two images with similar objects. The lines between these two
images represent the corresponding keypoints found on both images. The rectangle
on the observed image pinpoints the object location (Fig. 2.6).
2.2 Edge Detection
Edge detection [7] is one of the most commonly used stages in image processing [30].
This step is extremely relevant in the image description process. It allows detecting
the object shape, which is crucial in further steps of image classification or retrieval.
Usually, edges are detected between two different regions in the image (gradient
change). They occur when the image intensity or its first derivative changes. The
visual representation of derivatives is presented in Fig. 2.7.
2.2 Edge Detection 13
Fig. 2.7 Examples of image derivatives
There are four types of edges (see Fig. 2.8) [30]

• Ramp—edge values increase gradually. This edge profile is commonly used in
photos. It can be also found after image blur process,
• Step—intensity value changes abruptly, from one side to another. This is the most
desirable edge profile. They are often observed in computer images,
• Line—this edge profile is composed of two closely positioned edges. They are
detected by the first derivative filters and they are immune to most blur techniques,
• Roof—this profile is frequently observed in lines on the photographic images.
Essentially, the edge detection process is based on detecting significant local changes
in the image. Let us take for consideration one dimension, the step edge is related to
a local peak of the first derivative. In such case, the gradient is defined as the measure
of change of a function. It can also be considered as an array of continuous function
Fig. 2.8 One-dimensional edge profiles
samples of image intensity. Thus, the gradient is equivalent of the first derivative and
it is described by [30]
⎡ ⎤ ⎡∂ f ⎤
Gx ∂x
G[ f (x, y)] = ⎣ ⎦=⎣ ⎦, (2.7)
∂f
Gy ∂y
where G x and G y is expressed by filters. In order to approximate the gradient in both

directions (x, y) the following filters can be used [2, 52]
• the Sobel operator ⎡ ⎤
−1 0 1
Gx = ⎣−2 0 2⎦ , (2.8)
−1 0 1
⎡ ⎤
1 2 1
Gy = ⎣ 0 0 0 ⎦ , (2.9)
−1 −2 −1
• the Prewit filter ⎡ ⎤

−1 0 1
Gx = ⎣−1 0 1⎦ , (2.10)
−1 0 1
⎡ ⎤
1 1 1
Gy = ⎣ 0 0 0 ⎦ , (2.11)
−1 −1 −1
• the Robert’s Cross operator

0 1
Gx = , (2.12)
−1 0
1 0
Gy = , (2.13)
0 −1
• the Scharr’s operator ⎡ ⎤

−3 0 3
Gx = ⎣−10 0 10⎦ , (2.14)
−3 0 3
⎡ ⎤
3 10 3
Gy = ⎣ 0 0 0 ⎦ . (2.15)
−3 −10 −3
In order to determine the edge strengths (the gradient magnitude), we can use the
Euclidean (2.16) distance measure [38]

|G| = G 2x + G 2y . (2.16)
It is common practice to approximate this magnitude by the absolute values

|G| ≈ |G x | + G y , (2.17)
or
|G| ≈ max(|G x | , G y ), (2.18)
where G x is horizontal direction gradient and G y is vertical direction gradient. The

edge direction (angle) is determined by the following formula [7, 38]
Fig. 2.9 The edge detection by the Sobel filters. Figure 2.9a—input image, Fig. 2.9b—edge detec-
tion (input image taken from the PASCAL VOC dataset [16])

Gx
Θ = arctan , (2.19)
Gy
where Θ is measured with respect to the x axis. It should be noted, that the gradient
magnitude is independent of the direction of the edge. An example of edge detection
by the Sobel filters is presented in Fig. 2.9.
2.2.1 Canny Edge Detection
Canny edge detection is one of the most commonly used image processing methods
for detecting edges [3, 59], proposed by John F. Canny in 1986. It takes as input a
gray scale image, and produces an image showing the positions of tracked intensity
discontinuities. The algorithm is composed of 5 separate steps:
1. Noise reduction. The image input is smoothed by applying an appropriate Gaus-
sian filter.
2. Finding the intensity gradient of the image. During this step, the edges are marked
where gradients of the image have large magnitudes.
3. Non-maxima suppression. If the gradient magnitude at a pixel is larger than
those at its two neighbours in the gradient direction, mark the pixel as an edge.
Otherwise, mark the pixel as the background.
4. Edge tracking by hysteresis. The final edges are determined by suppressing all
edges that are not connected to the genuine edges.
The result of the Canny operator is determined by the following parameters:
• Width—of the Gaussian filter used in the first stage directly affects the results of
the Canny algorithm,
• Threshold (step)—used during edge tracking by hysteresis. It is difficult to provide
a generic threshold that works well on all images.
The Canny detector finds edges where the pixel intensity changes (image gradient).
Before edge detection, not important edges need to be removed. Thus, the Gaus-
Fig. 2.10 Figure 2.10a shows an input image and Fig. 2.10b represents the edge detected image.
As shown, the edges were detected correctly, because the image gradient is low (Fig. 2.10a was
taken from the Corel image dataset [55])
Fig. 2.11 Another example of edge detection on an image taken from the Corel dataset [55]
sian smoothing is applied. This process is performed by the Gaussian function for
calculating transformation for each image pixel [38]
1 x 2 +y 2
G(x, y) = √ e− 2σ2 , (2.20)
2πσ 2
where x is the distance from pixel position in the horizontal axis, y is the distance from
pixel position in the vertical axis, σ is the standard deviation of Gaussian distribution
(Fig. 2.11).
In some cases, simple edge detection does not provide satisfying results. The
edges are not always complete, and they do not describe the entire object. In order
Fig. 2.12 The edge linking process. Figure 2.12a—input image, Fig. 2.12b—edge detection, Fig.
2.12c—edge linking
to eliminate this nuisance, edge linking method is used. This algorithm simply con-
nects the edges in close proximity. This process is performed by repeating the edge
detection, with a lower step value parameter. The entire process is presented in
Fig. 2.12.
2.3 Blob Detection and Blob Extraction
Blob detection is one of the commonly used methods in image processing. It allows to
detect and extract a list of blobs (objects) in the image. In most cases, the blob distin-
guishes from the background and the other object. The filters used in edge detection
cannot be used due to the large size of the objects. The most common approach used
as preprocessing in blob detection is applying the Gaussian blur. Thus, the obtained
image is simplified, and not important details are removed. The blob detection is
widely used in various areas such as medicine [44], military, forensics [63] and
many others. Unfortunately, obtaining homogeneous objects from an image as a list
of pixels is extremely complicated. Especially when we deal with a heterogeneous
background, i.e. the objects containing multicoloured background. There are many
methods for extracting objects (blobs) from images [13, 22, 24]. These algorithms
are described by Andrew Kirillov [34]. The author distinguished four types of the
filters: Convex full, Left/Right Edges, Top/Bottom Edges, Quadrilateral. Figure 2.13
presents these blob detection methods. Figure 2.13a illustrates Convex full filter. As
can be seen, round edges of the objects are not detected correctly. Much better results
are obtained by the Top/Bottom Edges filter (Fig. 2.13C). Object edges are detected
mostly correctly, with individual exceptions. The Left/Right Edges method behaves
similarly (Fig. 2.13b). The last method has a problem with the detection of vertices
inside figures, e.g. star-shaped objects (Fig. 2.13d).
Blob extraction can be described by the following steps (Quadrilateral):
1. Locate each separate object in the input image,
2. Find object edge pixels (methods: Top/Bottom, Left/Right),
3. Detect four corners of the quadrilateral,
2.3 Blob Detection and Blob Extraction 19
Fig. 2.13 Comparison of methods for blob detection used in the AForge.NET library [34]
4. Set distor tion Limit. This value determines how different the object can be. Set
0 to detect perfect quadrilateral,
5. Check how well the analyzed shape fits into the quadrilateral with the assumed
parameter (see Fig. 2.13d).
6. Check each mean distance between given edge pixels and the edge of assumed
quadrilateral,
7. If the mean distance is not greater than distor tion Limit, then we can assume
that the shape is quadrilateral.
2.4 Clustering Algorithms
Data clustering is widely used in data mining and its applications. Clustering is a
process of organising data sets into homogeneous groups. This organisation process is
based on the given features of similar objects. There were many clustering algorithms
proposed so far.
2.4.1 K -means Clustering Algorithm
K -means algorithm belongs to the group of heuristic clustering algorithms. It was

firstly proposed by MacQueen in 1967 [40] and improved by Hartigan in 1979 [28].
The purpose of this method is to divide N points in D dimensional space into K clus-
ters. That process of partitioning requires only one input parameter K the number of
clusters [19]. K -means is commonly used for data clustering in various applications.
It is relatively fast and easy in implementation. The following pseudo-code describes
its functioning [32, 58]:
INPUT: X = {x1 , x2 , ..., xn } ∈ R n (data to cluster), k (number of clusters)
OUTPUT: C E N = {cen 1 , cen 2 , ..., cen n } ∈ R n (cluster centers),
L = {l1 , l2 , ..., ln }
for i = 1; i <= n; i + + do
C E N [i] = rand();
end
foreach xi ∈ X do
li := Ar g Min Dist (xi , cen j ), j ∈ {1..k}
end
ch := f alse;
while ch = f alse do
foreach cen i ∈ C E N do
U pdateCluster (cen i );
end
foreach xi ∈ X do
min Distance := Ar g Min Dist (xi , cen j ), j ∈ {1..k}
if min Distanceli then
li := min Distance
ch := tr ue
end
end
end
Algorithm 1: K -means algorithm.
Figure 2.14 shows the effect of colour clustering, where the left image is the input and
the right one is the output. As can be seen, the number of colours has been reduced
and only important features remain.
2.4.2 Mean Shift Clustering Algorithm
Mean shift is a clustering algorithm which does not require setting the number of
output clusters or their shapes. The number of parameters of the algorithm is in
fact, limited to the radius [11]. The basic version of the algorithm was presented in
the two-dimensional Euclidean space. The task of the algorithm is to compute the
2.4 Clustering Algorithms 21
Fig. 2.14 Image colour clustering by the k-means algorithm
Euclidean distance between each point and all other points. Mean shift determines
the points in d-dimensional space as a probability density function, where the denser
regions correspond to local maxima. For each point in this space, there is performed a
procedure of gradient increase until coverage. Points assigned to one agent of group
(stationary point) are considered to be a part of the group and form a single cluster
(group) [17]. Given n points xi ∈ R d , multivariate kernel density function K (x) is
expressed using the following equation [10, 11, 17]
n
1 x − xi
fk = , (2.21)
nh d i=1 h
where h is a radius of the kernel function. The kernel function is defined as follows
[10, 17] K (x) = ck k( x 2 ), where ck is a normalization constant. If estimator
density gradient is given it is possible to make the following calculations [14]
n n
2c x − x xi g x−x i
2
∇
k,d i
f (x) = g 2 i=1
n x−xi − x ,
h
(2.22)
nh d+2 i=1 i=1 g h
h 2
ter m1 ter m2
where g(x) = −k (x) is derivative of selected kernel of function. First term (ter m1)
of formula 2.22 allows to define the density and second term (ter m2) is named as
mean shift vector m(x). Points in the direction of maximum gain and proportional
to the density gradient can be determined at the point x obtained with the kernel
function K .
The algorithm can be represented in the following steps [5, 14, 39]:
1. Determine mean shift vector, expressed by the formula m(xti ),
2. Map the estimated density of window: xt+1 i
= xti + m(xti ),
3. Repeat the first and the second step until: ∇ f (xi ) = 0.
Then the group measures are calculated, and they are assigned to the selected group.
This process is presented in Fig. 2.15. Figure 2.15a presents an example of the points
arranged in two-dimensional space. Figure 2.15b presents the results of the algorithm
and the detected six different classes.
Another random document with
no related content on Scribd:
“‘No, you don’t, old hayseed; hand it here.’ I was so taken aback at
his calling me hayseed—you see, Modeste, I had on my blouse,”
and Paichoux looked a little guilty while referring to his toilet.
“Well papa, haven’t I told you not to go uptown in your blouse?”
said Tante Modeste sharply. “I should think now, for Marie’s sake,
that you would wear a coat; the Guiots all wear coats.”
“Oh, never mind that. I don’t. I’m an honest man, and I can afford
to wear a blouse anywhere. I didn’t take any notice of his
impudence, but I offered him a hundred. You see I happened to have
the money with me. I was on my way to pay Lenotre for those last
Jerseys I bought from him, so I took my wallet out and began
counting the bills. That brought him; the fellow needed the money,
and he wanted to get rid of the watch. If I hadn’t thought that there
was something crooked about it, my conscience wouldn’t have let
me take such a valuable thing for such a price, but I considered the
child. I thought it might be all the proof that we would ever have if
anything came up, and in any case it’s money well invested for her.”
“You did right to buy it, Paichoux. It’s a good deal of money for a
watch, especially just now, when we have to get so much for Marie;
but if we can do anything for that darling by having it, I don’t mind.”
And Tante Modeste sat for some time looking intently at the
beautiful, sparkling object that lay on her white apron.
“I wish it could speak,” she said at length; “I wish it could speak.”
“I mean to make it by and by,” returned Paichoux decidedly.
“But now, at this moment, what a story it could tell if it had a voice!
Well, I’m glad we’ve got it out of that scamp’s clutches.”
“So am I,” returned Paichoux, opening the case as he spoke and
showing Tante Modeste something on the inside of it. “I can get a
trace through this, or I’m mistaken; but put it away now in my safe,
and say nothing about it,—I don’t want even Madelon to know that
we’ve got it, and, Modeste, whenever you see that woman, be on the
alert for something that will give us a clue.”
“Oh, Paichoux, you don’t know her. She’s as close as the grave,
and too cunning to betray herself. I’m always watching her, and I
mean to keep on; but I don’t think it’s any use. I wish we could
employ a detective to unravel the mystery.”
“Yes, yes; but that would cost a good deal, Modeste; let’s wait
awhile, something’s going to turn up to put us on the right track.”
“And in the mean while the poor little darling is in the power of that
woman. The child never complains, but my heart aches for her. She
has changed this summer; she looks thin and weak, and that woman
takes no more care of her than she would of a dog. If it wasn’t for
Madelon and Pepsie, and Mam’selle d’Hautreve, the little creature
would suffer; and our good milk that I send to Madelon has helped
her through the hot weather. Pepsie herself goes without, to give it to
the child. If the sweet little thing hadn’t made friends, she would have
perished.”
“Let her come down here and play with our young ones; there’s
room enough,” said Paichoux good-naturedly, “and she’s no more
trouble than a bird hopping about.”
“I wanted to have her, but madame won’t let her come; she’s taken
it in her head to keep the child shut up most of the time. Pepsie and
Mam’selle Diane complain that they don’t have her as often as they’d
like to. I think she’s afraid that the child may talk. You see she’s
getting older, and she may remember more than madame likes her
to.”
“Well,” said Paichoux deliberately, “I’ve made a plan, and by and
by I’m going to put it in operation. Just keep quiet and wait until I’m
ready to put my plan in operation.”
And Tante Modeste promised to wait.
CHAPTER XXIII
MADAME JOZAIN CALLS UPON MAM’SELLE DIANE
I T was somewhere about the time that Paichoux bought the watch
when Mam’selle Diane was surprised one morning by a visit from
Madame Jozain, who entered the little green gate with an air of
haughty severity and insolent patronage that was insufferable; and
she had evidently come on business, for, after the first formalities
had passed between them, she drew a well-filled purse from her
pocket and asked in a lofty tone if Mam’selle Diane had her bill
prepared.
“My bill, Madame Jozain? What bill?” said Mam’selle Diane,
looking at her with cold surprise. “I am not aware that you owe me
anything.”
“I owe you for teaching Lady Jane music; you’ve been giving her
lessons now for some months, and I’m sure you must need your
money.”
“Oh, Madame,” gasped Mam’selle Diane, “you are laboring under
a mistake. I never thought of receiving money for the pleasure I have
had with the child. I offered to teach her. It was my own offer. You
surely did not think that I expected to be paid?”
“I certainly did. Why should you teach her for nothing when I am
able to pay?” returned madame haughtily, while she fingered her roll
of notes. “In your circumstances you can’t afford to throw away your
time, and I’m quite willing to pay you the usual price. You’re a very
good teacher, and I’m very well satisfied with the child’s progress.”
For a moment, Mam’selle Diane was quite overcome by the
woman’s insolence. Then, remembering that she was a d’Hautreve,
she drew herself up, and said calmly and without the least hauteur, “I
regret, Madame, that you thought me a teacher of music. I make no
claim to any professional knowledge, therefore I could not take the
pay of a teacher. I thank you very much, but I am not a teacher.”
“It doesn’t matter. I insist on paying you.” And madame held out a
bank-note for such a large amount that Mam’selle Diane’s eyes were
fairly dazzled.
“I assure you it is impossible,” said Diane gently. “It is useless to
discuss the matter. Will you permit me to open the gate for you?”
“Very well, then,” exclaimed madame, hotly. “I sha’n’t allow my
niece to come here again. I won’t accept favors from any one. She
shall have a teacher that isn’t too proud to take pay.”
“I hope you will not deprive us of the pleasure of seeing Lady
Jane. We are very fond of her,” said Mam’selle Diane, almost
humbly, while the tears gathered on her eyelashes. “Of course you
must do as you think best about the lessons.”
“I sha’n’t allow her to run about the neighborhood any more,”
replied madame, tartly; “she’s losing her pretty manners. I shall keep
her with me in the future,” and with this small parting thrust and a
curt good-morning she went out of the little green gate, and left
Mam’selle Diane to close it behind her with a very heavy heart.
The interview had taken place on the gallery, and Madame
d’Hautreve had heard but little from her bed. “Diane, what did that
woman want? What sent her here at this hour?” quavered the old
lady sharply.
“She came on business, mama,” replied Mam’selle Diane,
brushing away a tear.
“Business, business; I hope you have no business with her.”
“She pretended to think I expected to be paid for the lessons I
have given Lady Jane.”
Madame groaned. “I told you we would regret opening our doors
to that child.”
“Oh, mama, I don’t regret it. I only regret that I have lost the
pleasure of seeing her. Madame Jozain will not allow her to come
any more.”
“Ungrateful creature, to insult you after your condescension.”
“Mama, she didn’t insult me,” interrupted Mam’selle Diane,
proudly. “Must I remind you that I am above her insolence?”
“True, my dear, true, and I hope you made her feel that she is a
Jozain.”
“I didn’t wish to be unkind to her, mama. Perhaps she is not so
wrong after all. Sometimes I think it would have been better to have
let our friends know our real circumstances. Then they would have
helped me to get pupils, and I could have earned more teaching
music than I can making pen-wipers, and I am sure it would be more
respectable and more agreeable.”
“Oh, Diane, you surprise me,” cried Madame d’Hautreve,
tremulously. “Think of it! a granddaughter of the Counts d’Hautreve
and d’Orgenois teaching the children of grocers and bakers to play
the piano. No, no; I would rather bury myself here and die in poverty
than disgrace our name in that way.”
Mam’selle Diane made no reply, and after a few moments
madame turned on her pillow to finish her morning nap. Then the last
of the d’Hautreves went into the little garden, and drawing on a pair
of old gloves she dug and trimmed and trained her flowers for some
time, and afterwards gathered up the small piles of seeds from the
white papers.
“Oh, oh!” she said wearily, seeing how few they were, “even the
flowers refuse to seed this year.”
After she had finished her work in the garden, she went dejectedly
back to the little room where her mother still slept, and opening a
drawer in her armoire she took out a small box. She sighed heavily
as she raised the lid. Inside on a blue velvet lining lay a slender
bracelet set with turquoises and diamonds. “It must go,” she said
sadly to herself. “I have kept it till the last. I hoped I wouldn’t be
obliged to part with it, but I must. I can’t let poor mama know how
needy we are. It’s the only thing I can spare without telling her. Yes, I
must give it up. I must ask Madame Jourdain to dispose of it for me.”
Then she sat for a long time looking at it silently, while the hot tears
fell on the blue velvet. At last, with a sigh, she bravely wiped her
eyes, and laid the little box under the ducklings in the black basket.
For more than a week Mam’selle Diane did not see Lady Jane,
and the poor woman’s eyes had a suspicious look of tears, as she
went about her duties, silent and dejected. Her only pleasure was no
longer a pleasure; she could not go near the piano for some days. At
last, one evening, she sat down and began to play and sing a little
song she had taught the child, when suddenly she heard, outside the
window, the sweet treble voice she loved so well.
“It’s Lady Jane!” she cried, and springing up so hastily that she
upset the piano-stool she grappled with the rusty bolts of the
shutters, and, for the first time in years, threw them boldly open, and
there stood the child, hugging her bird to her breast, her wan little
face lit up with her sparkling eyes and bright, winsome smile.
Mam’selle Diane went down on her knees, and Lady Jane clung to
her neck and kissed her rapturously over and over.
“Diane, Diane, what are you thinking of, to open that shutter in the
face of every one?” said the old lady feebly.
But Mam’selle Diane did not hear her mother; she was in an
ecstasy of happiness, with the child’s soft lips pressed to her faded
cheeks.
“Tante Pauline says I mustn’t come in,” whispered Lady Jane,
between her kisses, “and I must mind what she says.”
“Yes, darling, you must obey her.”
“I’ve been here every day listening, and I haven’t heard you sing
before.”
“Dear child, I couldn’t sing; I missed you so I couldn’t sing.”
“Don’t cry, Mam’selle Diane; I love you dearly. Don’t cry, and I’ll
come every day to the window. Tante Pauline won’t be angry at that.”
“I don’t know, my dear; I’m afraid she will.”
“Diane, close that window instantly,” cried Madame d’Hautreve,
quite beside herself. “A pretty exhibition you’re making before all the
neighbors, on your knees crying over that child.”
“Good-by, darling; come sometimes. Mama don’t like me to open
the window, but I’ll open the gate and speak to you,” said Diane,
hastily returning to herself and the exigencies of her position.
“Forgive me, mama, I really couldn’t help it, I was so glad to see
the child,” and Mam’selle Diane closed the window with a brighter
face than she had shown for several days.
“I think you must be insane, Diane, I surely think you must be, to
let all these common people know that a blanchisseuse de fin will not
allow her child to come into our house, and that you are obliged to
go on your knees and reach out of the window to embrace her. Oh,
Diane, Diane, for the first time you’ve forgotten that you’re a
d’Hautreve!”
CHAPTER XXIV
RASTE THE PRODIGAL
A BOUT this time, a noticeable change took place in Madame

Jozain. She did not seem nearly so self-satisfied, nor so
agreeable to her customers. They remarked among themselves that
something had certainly gone wrong, for madame was very absent-
minded and rather cross, and was always talking about business
being poor, and the quarter growing duller every day, while the
neighbors were a set of curious gossips and busybodies.
“As soon as they find out that one has had trouble, they blacken
one all they can,” she said bitterly to Madame Fernandez, who was
her only intimate friend.
She spoke cautiously and vaguely of her troubles, for she did not
know whether the news of Raste’s escapade had reached Good
Children Street or not. “I dare say they have seen it in the papers,”
she thought angrily to herself. “Locked up for thirty days, as a
suspicious character! If he had listened to me, and sold that watch at
first, he wouldn’t have got into this trouble. I told him to be careful,
but he was always so headstrong, and now, I don’t know what may
happen any moment. The whole story may get out, through that
watch being talked about in the papers, and perhaps the man that
bought it was a detective. Raste didn’t even find out who bought it. I
shall never feel easy now until Raste is out of the way. As soon as
thirty days are ended, I shall advise him to leave New Orleans for a
while. I’m disgusted with him, to disgrace me in this way, and I don’t
want him here. I can hardly make enough to support myself and that
child. If it wasn’t for the money I’ve hidden away, I should feel
discouraged; but I’ve got that to fall back on. I’m thankful Raste don’t
know anything about it, or he’d get it from me in some way. I’m glad
I’ve got rid of all those things; I’d be afraid to have them by me now.
There’s nothing of any consequence left but that silver jewel-box,
and I’ll get that off my hands the first time I go out.”
Then she thought of the child. Suppose some one should
recognize the child? She was becoming cowardly. A guilty
conscience was an uncomfortable companion. Everything frightened
her and made her suspicious. Madame Paichoux had asked some
startling questions; and besides, she did not know what the child
might tell. Children were so unreliable. One would think they had
forgotten everything and did not see nor hear; then, suddenly, they
would drop some word that would lead to wonderful revelations.
Lady Jane was an intelligent, thoughtful child, and such people as
the d’Hautreves could find out many things from her. Then she
congratulated herself that she had been clever enough to get her
away from Mam’selle Diane, and the Paichoux, too. And that
cunning little hunchback, Pepsie; and old Gex—he was a sly old
villain, and no doubt her enemy, for all he was so affable and polite.
Yes, she would keep the child away from them all as much as
possible.
Sometimes she thought it would be best to move away from that
quarter of the city; but then, her going might excite suspicion, so she
waited for further developments with much anxiety.
When Raste’s thirty days were up he came to his mother, very
sheepish, and, apparently, very penitent. To her angry reproaches,
he replied that he had done nothing; that there was no crime in his
having the watch. They didn’t steal the watch; they didn’t ask the
poor woman into their house and rob her. She came there sick, and
they took care of her; and instead of turning her child into the street,
they had treated her as if she belonged to them. As for the watch, he
had been keeping it only until the child was old enough to have it, or
until her relatives turned up; he had never intended to sell it until he
found that it was getting him into trouble, and then he was obliged to
get rid of it.
Madame listened to the plausible arguments of her handsome
scapegrace, and thought that perhaps, after all, there was no real
cause for anxiety; and when he treated his thirty days with fine
scorn, as a mere trifle, a mistake of which no one knew, she felt
greatly comforted.
“Respectable people,” he said, “never read about such matters,
and, consequently, none of our friends will ever know of it. It won’t
happen again, for I mean to cut loose from the fellows who led me
into that fix. I mean to go with respectable people. I shall begin all
over, and earn a living in an honest way.”
Madame was delighted; she never knew Raste to talk so
reasonably and to be so thoughtful. After all, his punishment hadn’t
done him any harm. He had had time to think, and these good
resolves were the result of his seclusion from the friends who had
nearly proved his ruin. Therefore, greatly relieved of her anxieties,
she took the prodigal back into her heart and home, and cooked him
an excellent supper, not of a fatted calf, but of a fatted pig that
Madame Paichoux had sent her as a preliminary offering toward
closer acquaintance.
For several days Raste remained quietly at work around the
house, assisting his mother in various ways, and showing such a
helpful and kindly disposition that madame was more than ever
enchanted with him. She even went so far as to propose that they
should form a partnership and extend their business.
“My credit is good,” said madame, proudly; “I can buy a larger
stock, and we might hire the store on the corner, and add a grocery
department, by and by.”
“But the capital? We haven’t the capital,” returned Raste
doubtfully.
“Oh, I’ll provide the capital, or the credit, which is just as good,”
replied madame, with the air of a millionaire.
“Well,” said Raste, “you go out among the merchants and see
what you can do, and I’ll stay here and wait on the customers.
There’s nothing like getting used to it, you know. But send that young
one over to the ‘Countess,’ or to some of her swell friends. I don’t
want to be bothered with her everlasting questions. Did you ever see
such a little monkey, sitting up holding that long-legged bird, and
asking a fellow a lot of hard questions, as serious as old Father
Ducros himself? By the way, I saw Father Ducros; he’s just back
from Cuba. I met him yesterday, and he asked me why you didn’t
come to church.”
Madame went out to see about the new venture with Father
Ducros’s name ringing in her ears, and was absent for several hours.
When she returned she found the house closed and Raste gone.
In a moment Lady Jane came running with the key. Mr. Raste had
brought it to her, and had told her that he was tired tending shop,
and was going for a walk.
Madame smiled, and said as she took the key:
“I thought so; I thought he’d get tired of it, but I can’t expect him to
keep closely to business just at first.”
She took off her bonnet and veil, and put them away; then went
limping about the room, putting it in order. From time to time she
smiled. She had met Madame Paichoux and Marie in the Bon
Marché on Rue Royale, and they had been very agreeable. Madame
Paichoux had even invited her to come and dine with them, to meet
Marie’s fiancé. At last they were beginning to see that she was
worthy of some attention, she thought.
Now, if Raste would only behave himself, they could do very well.
With the ready money she had hidden away and by using her credit
she could buy a large stock of goods. She would have more shelves
put up, and a counter, and a fine show-case in the window; and there
was the store on the corner which Raste could fit up as a grocery.
Suddenly she remembered that her rent was due, and that it was
about time for her landlord’s visit. She took out her pocket book and
counted its contents. She had been rather extravagant at the Bon
Marché, to impress Madame Paichoux, and had spent far more than
she intended. She found that she lacked a few dollars of the amount
due for rent.
“I must borrow it from the private bank,” she said jocosely, as she
unlocked her bureau.
With the peculiar slyness of such people, she thought her hoard
safer when not too securely concealed. Therefore she had folded up
the whole of her year’s savings, with the amount taken from Lady
Jane’s mother, inside of a pair of partly worn gloves, which were
thrown carelessly among her other clothing in the drawer. It was true
she always kept her bureau locked, and the key was hidden, and
she seldom left her house alone. But even if any one should break it
open, she thought they would never think of unrolling those old
gloves.
When she opened the bureau it seemed very disorderly. “I didn’t
surely leave my things in such confusion,” she said, nervously
clutching at the gloves, which were startlingly conspicuous. With
trembling hands and beating heart she unfolded them, but instead of
the roll of notes only a slip of paper was found.
The gloves dropped from her nerveless fingers, and, staggering to
her bed, she sat down on the edge and read the large characters,
which were only too familiar and distinct, although they danced and
wavered before her eyes:
Dear Mama:
I’ve decided not to go into partnership with you, so I’ll take the
capital and you can keep the credit. The next time that you secrete
from your dutiful son money that you have no right to, don’t hide it in
your old gloves. It isn’t safe. I’m going away on a little trip. I need a
change after my close application to business. By the way, you can
tell your inquisitive neighbors that I’ve gone out to my uncle’s ranch in
Texas.
Your affectionate and devoted son,
Adraste Jozain.
CHAPTER XXV
THE JEWEL-BOX
T HE next day after Raste’s sudden departure, Madame Jozain sat

in her doorway looking very old and worn; her face was of a
settled pallor, and her eyes had a dazed, bewildered expression, as
if she had received a heavy blow that had left her numb and stupid.
At times she put her hand to her head and muttered, “Who would
have thought it? Who would have thought it? His mother, his own
mother, and I’ve always been so good to him.”
Suddenly she seemed to have lost her interest in her business,
her customers, and even her domestic affairs. Her little store was
more untidy than any one had ever seen it. When a neighbor entered
to buy a trifle, or to gossip for a few moments, madame made an
effort to appear cheerful and chatty, but that it was an effort was
evident to all. At last some one asked if she were ill.
“Well, not exactly,” she answered uneasily, “but I might as well be.
The fact is I’m fretting about that boy of mine; he took it in his head
yesterday to go away to his uncle’s ranch. I miss him very much. I
can’t get along without him, and I shouldn’t wonder if I should go
too.”
When Pepsie asked what was the matter with “Tante Pauline,”
Lady Jane answered, as she had been instructed, that Tante Pauline
had headaches, because Mr. Raste had gone away and wasn’t
coming home for a long time.
“Madame Jozain is fretting about her son’s going away,” observed
Madame Fernandez to her husband, looking across the street.
“She’s been sitting there all the morning so lonesome and miserable
that I’m sorry for her. But there’s some one coming to see her now. A
stranger, and so well-dressed. I wonder who it can be.”
The new-comer was a stranger to Madame Fernandez, but
Madame Jozain welcomed her as an old friend; she sprang up with
sudden animation and shook hands warmly.
“Why, Madame Hortense,” she exclaimed, “what chance brings
you to my little place?”
“A happy chance for you,” replied Madame Hortense, laughing.
“I’ve come to bring you money. I’ve sold the little jewel-case you left
with me the other day, and sold it very well, too.”
“Now, did you? How good of you, my dear! I’m so glad—for the
child’s sake.”
“Would you believe that I got twenty-five dollars for it? You know
you said I might sell it for ten; but I got twenty-five, and I think I could
have sold it for more easily. It is solid silver and an exquisite thing.”
“Yes, it was of the best workmanship,” sighed madame.
“But I must tell you how I happened to sell it for such a high price.
It’s very strange, and perhaps you can throw some light on the
matter. One of my best customers happened to come in last evening
—Mrs. Lanier, of Jackson Street. You know Lanier, the banker. They
are very rich people. She was looking over the things in my show-
case, when she suddenly exclaimed as if surprised:
“Why, Madame Hortense, where did you get this?” I turned
around, and she had the little jewel-box in her hand, examining it
closely, and I saw that she was quite pale and excited.
“Of course I told her all I knew about it; that a friend had given it to
me to sell, and so on. But she interrupted me by asking where my
friend got it, and all sorts of questions; and all the while, she was
looking at it as if she couldn’t imagine how it got there. I could only
tell her that you gave it to me. Then she asked other questions, so
excitedly that I couldn’t help showing my surprise. But I couldn’t give
her the information she wanted, so I wrote your name and address
for her, and told her to come and see you, and that you would be
able to tell her all about it.”
During Madame Hortense’s hasty and rather confused narrative
Madame Jozain turned an ashy white; and her eyes took on a
hunted expression, while she followed with a set, ghastly smile every
word of her friend’s story.
At length she found strength and composure to say:
“Why, no wonder you were surprised. Didn’t she tell you why she
wanted to know?”
“I suppose she saw that I was very much puzzled, for after looking
at it sadly for some time, she said that it was a mystery how it came
there; that she had given that little casket to a schoolmate ten years
before, while at school in New York; that she had had it made
especially for her, and that her friend’s initials, J. C., were on it.”
“Dear, dear, only think! Some old schoolmate, I suppose,” said
Madame Jozain hastily.
“Then she asked me if I would sell her the little box; and I said
certainly I would, that it was put there to sell. Seeing how anxious
she was to get it, I thought I would put the price at twenty-five
dollars, although I didn’t much think she’d give it. But she never said
a word about the price; she paid it in a dazed way, took your address
that I’d written for her, and went out, carrying the little casket with
her. I suppose she’ll be here to-day or to-morrow to see you; and so I
thought I’d hurry down and tell you all about it.”
“And your commission?” said Madame Jozain, with a visible effort,
as the milliner laid the money on the table.
“Oh, par exemple, Madame Jozain! As if I would! No, no; we’re too
old friends. I cannot take pay for doing you a little favor. And besides,
I’m glad to do it for the dear child. She must be a great anxiety to
you.”
“She is!” returned madame, with a heavy sigh, “but she has some
property in Texas, I believe. My son has just gone there, and I’m
thinking of going too. I’m very lonely here.”
“Ah!” said Madame Hortense, surprised. “Why, you are so well
placed here. Shall you go soon?”
“Before very long,” replied madame, who did not care to be more
definite.
“Well, come and see me before you go.”
Madame Hortense drew down her veil, and rose to leave. “I’m
sorry I can’t stay longer to chat with you; I’m busy, very busy. Now
mind, be sure to come and say good-by,” and with a cordial au revoir
the little milliner hurried down the steps and out of sight around the
corner.
For some time after her visitor had gone, Madame Jozain stood
quite still in the middle of her little shop, with her hands pressed to
her head and her eyes fixed on vacancy. At length she muttered to
herself: “She’ll come here; yes, she’ll come here! I can’t see her; I
can’t tell her where I got that box. I must get away at once. I must go
out and find another place. There’ll be no more peace on earth for
me! My punishment’s begun!”
Then madame hurriedly put on her best gown and bonnet, and
calling across to Lady Jane, who was with Pepsie, she said she was
going out on business, and that she might not be back for some
time.
CHAPTER XXVI
THE FLIGHT
L ATE that same afternoon, Madame Jozain was limping slowly

and wearily through a narrow street at the other end of the city,
miles away from Good Children Street, when she saw an old negro
sitting on a furniture-wagon to which two mules were harnessed.
“Is that you, Pete?” she asked, stopping and looking at him.
“Why, law, yes, it’s me, Miss Pauline, an’ I is mighty glad ter see
yer,” said the old man, climbing down.
“And I’m glad to find you, Pete. I see you’ve got a wagon. Is it
yours?”
“Well, ’t ain’t edzectly mine, Miss Pauline. I is hired it. But I is a-
drivin’ it.”
“I was just looking for some one to move me to-night, Pete.”
“Ter-night, Miss Pauline? Why, we doesn’t often work a’ter
sundown, an’ it’s mos’ dat now.”
“What do you charge for a load, Pete, when you move furniture?”
“I mos’ gen’ly charges two dollars a load—when it ain’t too fur,
Miss Pauline,” he answered slowly.
“Well it is far, Pete; it is from Good Children Street.”
“Oh, Miss Pauline, I can’t do dat dar ter-night. My mules is too
tired for dat.”
Madame stood still and thought for a moment.
“See here, Pete,” she said at length in a tone of decision; “I want
you to remember that you belonged to our family once, and I want
you to listen to me, and do what I tell you. You’re to ask no
questions, nor answer none; mind that! You’re to keep your tongue
still. Take your mules out now, and give them a good feed, and let
them rest awhile. Then be at my house by ten this evening. That will
be soon enough, for I’ve got to pack. If you’ll move me quietly, and
without any fuss, I’ll give you ten dollars for the load.”
“Ten dollars, Miss Pauline?” and the old darky grinned. “Bress yer,
miss, I is a mind ter try it—but it’s a mighty long road!”
MADAME JOZAIN BARGAINS FOR HER MOVING
“You’ve got plenty of time; you needn’t hurry. Bring a man to help,
and leave your wagon in the side street. I want the things taken out
the back way, and no noise. Mind what I say, no noise.”
“All right, Miss Pauline, I’ll be dar shore. An’ yer’ll gib me ten
dollars?”
“Yes, ten dollars,” replied madame, as she limped away to take the
street-car.
Some of Madame Jozain’s neighbors remembered afterward that
they slept badly that night—had uneasy dreams and heard
mysterious noises; but as there was a thunder-storm about
daybreak, they had concluded that it was the electricity in the air
which caused their restlessness. However, Pepsie afterward insisted
that she had heard Lady Jane cry out, and call “Pepsie!” as if in great
distress or fear, and that about the same time there were sounds of
hushed voices, rumbling of wheels, and other mysterious noises. But
her mother had told her she was dreaming.
So upset was Pepsie by the night’s experience that she looked
quite pale and ill as she sat by her window next morning, waiting for
Madame Jozain to open the shutters and doors.
How strange! It was eight o’clock, and still no sign of life in the
house opposite! The milkman had rung his bell in vain; the brick-dust
vender had set his bucket of powdered brick on the very steps, and
shrieked his discordant notes close to the door; the clothes-pole man
had sung his dismal song, and the snap-bean woman had chanted
her three syllables, not unmusically, and yet no one appeared to
open the door of Madame Jozain’s house.
At last Pepsie could endure her suspense no longer.
“You go and see what is the matter,” she said to her little
handmaid.
So Tite zigzagged across the street, flew up the steps, and
pounded vigorously on the door; then she tried the shutters and the
gate, and finally even climbed the fence, and peeped in at the black
windows. In a trice she was back, gasping and wild-eyed:
“Bress yer, Miss Peps’. W’at I done tol’ yer? Dem’s all gone. Ain’t a
stick or nofin’ in dat dar house! Jes’ ez empty ez a gourd!”
At first Pepsie would not believe the dreadful news; but finally,
when she was convinced that madame had fled in the night and
taken Lady Jane with her, she sank into the very depths of woe and
refused to be comforted.
Then Paichoux and Tante Modeste were called into a family
council, and Paichoux did his very best to solve the mystery. But all
he could learn was from madame’s landlord, who said that Madame
Jozain had paid her rent and given up her key, saying that she had
decided, very suddenly, to follow her son. This was all the
information the landlord could give, and Paichoux returned
dejectedly with this meager result.
“I had my plans,” he said, “and I was waiting for the right moment
to put them in operation. Now, the child has disappeared, and I can
do nothing.”
The next day Pepsie, sitting sorrowfully at her window, trying to
find consolation in a game of solitaire, saw a private carriage drive
up to the empty house and wait, while the servant made inquiries for
Madame Jozain.
“Madame Jozain did live there,” said M. Fernandez politely, “but
she went away between two days, and we know nothing at all about
her. There was something strange about it, or she never would have
left without telling her friends good-by, and leaving some future
address.”
The servant imparted this scanty information to the lady in the
carriage, who drove away looking greatly disappointed.
The arrival of this elegant visitor directly following upon madame’s
flight furnished a subject for romantic conjecture.
“I shouldn’t wonder,” said Pepsie, “if that was Lady’s mama, who
has come back after all! Oh, how dreadful that she wasn’t here to
see her!” and then poor Pepsie cried, and would not be consoled.

Download textbook Computer Vision Methods For Fast Image Classi%Ef%Ac%81Cation And Retrieval Rafal Scherer ebook all chapter pdf

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Download textbook Computer Vision Methods For Fast Image Classi%Ef%Ac%81Cation And Retrieval Rafal Scherer ebook all chapter pdf

Uploaded by

Copyright:

Available Formats

Computer Vision Methods for Fast

Image Classi■cation and Retrieval

Feature extraction and image processing for computer

Feature Extraction and Image Processing for Computer

Image Processing and Computer Vision in iOS Oge Marques

Nonlinear Eigenproblems in Image Processing and

Image Processing Computer Vision and Pattern

Computer Vision and Image Analysis Digital Image

Numerical Algorithms Methods for Computer Vision

Energy Minimization Methods in Computer Vision and

More information about this series at http://www.springer.com/series/7092

Computer Vision Methods

ISSN 1860-949X ISSN 1860-9503 (electronic)

Library of Congress Control Number: 2018968376

© Springer Nature Switzerland AG 2020

Częstochowa, Poland Rafał Scherer

3.3 Fast Two-Level Image Indexing . . . . . . . . . . . . . . . . . . . . . . . . . 55

5.2 Bag of Features Image Retrieval in Relational Databases . . . . . . . 114

© Springer Nature Switzerland AG 2020 1

2.1 Local Features

2.1.1 Scale-Invariant Feature Transform (SIFT)

SIFT (Scale-Invariant Feature Transform) is an algorithm used to detect and describe

© Springer Nature Switzerland AG 2020 7

L(x, y, σ) = G(x, y, σ) ∗ I (x, y), (2.1)

2.1.2 Speed-Up Robust Features (SURF)

Fig. 2.1 Calculating area

The determinant of this matrix can be calculated by [15]

Fig. 2.4 Visual

The descriptor consists of 4 × 4 wavelet group matrix. Each of them is composed of

Fig. 2.6 Another example of keypoint matching by the SURF algorithm

2.2 Edge Detection

Fig. 2.7 Examples of image derivatives

There are four types of edges (see Fig. 2.8) [30]

Fig. 2.8 One-dimensional edge profiles

where G x and G y is expressed by filters. In order to approximate the gradient in both

• the Prewit filter ⎡ ⎤

• the Robert’s Cross operator

• the Scharr’s operator ⎡ ⎤

It is common practice to approximate this magnitude by the absolute values

where G x is horizontal direction gradient and G y is vertical direction gradient. The

2.2.1 Canny Edge Detection

2.3 Blob Detection and Blob Extraction

2.4 Clustering Algorithms

2.4.1 K -means Clustering Algorithm

K -means algorithm belongs to the group of heuristic clustering algorithms. It was

2.4.2 Mean Shift Clustering Algorithm

Fig. 2.14 Image colour clustering by the k-means algorithm

A BOUT this time, a noticeable change took place in Madame

T HE next day after Raste’s sudden departure, Madame Jozain sat

L ATE that same afternoon, Madame Jozain was limping slowly

MADAME JOZAIN BARGAINS FOR HER MOVING

You might also like