IP seminar for Fall semester in 2004

IMAGE SEGMENTATION
Xiaoheng Yang Nakajima Lab, Titech October 29, 2004
CONTENTS

Problem Preview
Segmentation subdivides an image into it constitute s regions or objects. The level to which the subdivision is carried depends on the problem being solved. That is, segmentation should stop when the objects of interests in an application have been isolated. Image segmentation algorithms generally are based on one of two basic properties of intensity values: discontinuity and similarity.

Preview 1 Detection of Discontinuities

1.1 Point Detection 1.2 Line Detection 1.3 Edge Detection

Edge Linking and Boundary Detection

2.1 Local Processing 2.2 Global Processing via Hough Transform 2.3 Global Processing via Graph-Theoretic Techniques

Detection of Discontinuities

Thresholding
3.1 3.2 3.3 3.4 3.5 3.6 Foundation The Role of Illumination Basic Global Thresholding Basic Adaptive Threshloding Optimal Global and Adaptive Thresholding Use of Boundary Characteristics for Histogram Improvement and Local Thresholding 3.7 Thresholds Based on Several Variables

The most common way to look for discontinuities is to run a mask through the image.

R = wi zi
i

Region-Based Segmentation
4.1 Basic Formulation 4.2 Region Growing 4.3 Region Splitting and Merging

Segmentation by Morphological Watersheds

5.1 5.2 5.3 5.4 Basic Concepts Dam Construction Watershed Segmentation Algorithm The Use of Markers

Figure.1 1.1 Point Detection 1.2 Line Detection

| R | T

The Use of Motion In Segmentaion

6.1 Spatial Techniques 6.2 Frequency Domain Techniques

Summary References and Further Reading

Figure.2

IP seminar for Fall semester in 2004

1.3 Edge Detection A point is being an edge point if its two--dimensional first-order derivative is greater than a specified threshold. A set of such points that are connected according to a predefined criterion of connectedness is by definition an edge. In practice, optics, sampling, other image acquisition imperfections yield edges that are blurred. The slope of the ramp is inversely proportional to the degree of blurring in the edge. The thickness of edge is determined by the length of the ramp. First-order derivatives of a digital image are based on various approximations of the 2-D gradient. Second-order derivative is defined as digital approximations to the Laplacian of a 2-D function. Conclusion: The first derivative can be used to detect the presence of an edge at a point in an image. Similarly, the sign of the second derivative can be used to determine whether an edge pixel lies on the dark or light side of an edge (zero -crossing). Problem: Derivatives are sensitive to noise.

Edge Linking and Boundary Detection

Edge detection algorithms typically are followed by linking procedures to assemble edge pixels into meaningful edges. (for example, breaks caused by noise.)
2.1 Local Processing Criteria: the strength of the response of the gradient operator / the direction of the gradient vector A point in the predefined neighborhood is linked to the pixel if both magnitude and direction criteria are satisfied. 2.2 Global Processing via Hough Transform Computational attractiveness Approach based on the Hough transform is as follow: a. Compute the gradient of an image and threshold it to obtain a binary image. b. Specify subdivisions in the plane c. d. Examine the counts of the accumulator cells for high pixel concentrations. Examine the relationship between pixels in a chosen cell.

2.3 Global Processing via Graph-Theoretic Techniques This representation provides a rugged approach that performs well in the presence of noise. Some terms used here Graph G = ( N , U )

arc for each pair ( ni , n j )

successor / parent level cost

Figure.4 Figure.3

IP seminar for Fall semester in 2004

Figure.6. 3.5 Optimal Global and Adaptive Thresholding Figure.5 A method for estimating thresholds that produce the minimum average segmentation error PDF ( probability density function ) 3.6 Use of Boundary Characteristics for Histogram Improvement and Local Thresholding The chances of selecting a good threshold are enhanced considerably if the histogram peaks are tall, narrow, symmetric, and separated by deep valleys. One approach for improving the shape of histogram is to consider only those pixels that lie on or near the edges between objects and the background.

Thresholding
3.1 Foundation

T = T [ x, y, p( x, y ), f ( x, y)]
Classification for threshold: local, global, dynamic or adaptive 3.2 The Role of Illumination The image resulting from poor illumination could be quite difficult to segment. 3.3 Basic Global Thresholding The success of this method depends entirely on how well the histogram can be partitioned. Heuristic approach based on visual inspection of the histogram. 3.4 Basic Adaptive Threshloding Issues: how to subdivide the image and how to estimate the threshold for each resulting subimage

Figure.7

IP seminar for Fall semester in 2004

3.7 Thresholds Based on Several Variables Multispectral thresholding The concept of thresholding now becomes one of finding clusters of points in multi-dimension space. (In general, segmentation problems requiring multiple thresholds are best solved using region growing methods. ) 4

Region-Based Segmentation

The objective of segmentation is to partition an image into regions. 4.1 Basic Formulation 4.2 Region Growing Region growing is a procedure that groups pixels or subregions into larger regions based on predefined criteria. Seed region 4.3 Region Splitting and Merging

Figure.8

Segmentation by Morphological Watersheds

Figure.9 5.2 Dam Construction The simplest way to construct dams separating sets of binary points is to use morphological dilation.

Segmentation by watersheds embodies many of the concepts of the other three approaches and, as such, often produces more stable segmentation results, including continuous segmentation boundaries. 5.1 Basic Concepts The concept of watersheds is based on visualizing an image in three dimensions: two spatial coordinates versus gray levels.

IP seminar for Fall semester in 2004

A marker is a connected component belonging to an image.

Figure.12

Figure.13 Figure.10

6
5.3 Watershed Segmentation Algorithm

The Use of Motion In Segmentaion

6.1 Spatial Techniques

Figure.14 6.2 Frequency Domain Techniques

Reference: Digital Image Processing, Rafael C. Gonzalez & Richard E. Woods, second edition 2002, Prentice Hall
Figure.11 5.4 The Use of Markers Problem: oversegmentation