Curvature scale space representation : theory, applications, and MPEG-7 standardization /

MPEG-7 is the first international standard which contains a number of key techniques from Computer Vision and Image Processing. The Curvature Scale Space technique was selected as a contour shape descriptor for MPEG-7 after substantial and comprehensive testing, which demonstrated the superior perfo...

Full description

Bibliographic Details
Main Author:	Mokhtarian, Farzin
Corporate Author:	SpringerLink (Online service)
Other Authors:	Bober, Miroslaw Z., 1965-
Format:	eBook
Language:	English
Published:	Dordrecht ; Boston : Kluwer Academic Publishers, 2003.
Series:	Computational imaging and vision ; v. 25.
Subjects:	MPEG 7. Image processing > Digital techniques. Image analysis. Form perception. Beeldverwerking. Vormwaarneming. Bildverarbeitung. Formbeschreibung. Electronic books.
Online Access:	Connect to the full text of this electronic book

Table of Contents:

1. Multi-Scale Representations for Free-Form Planar Curves 1
1.2 The Curvature Scale Space Image 6
1.3 The Renormalized Curvature Scale Space Image 13
1.4 The Resampled Curvature Scale Space Image 16
1.5 Evolution and Arc Length Evolution Properties of Planar Curves 19
1.6 Experiments, Discussion, and Evaluation 26
2. Robust Free-Form Object Recognition Through Curvature Scale Space 33
2.2 Silhouette-Based Isolated Object Recognition 35
2.2.1 Curvature Scale Space Matching 36
2.2.2 A Silhouette-based Object Recognition System 38
2.2.3 Results and Discussion 40
2.3 Silhouette-Based Occluded Object Recognition 43
2.3.1 Image Segmentation 44
2.3.2 Multi-Scale Segmentation of 2-D Contours 44
2.3.3 Efficient Termination of Multi-Scale Segmentation 49
2.3.4 Local Matching through CSS 55
2.3.4.1 Rescaling 56
2.3.4.2 Candidate Generation and Filtering 57
2.3.4.3 Candidate Merging 58
2.3.4.4 Candidate Extension 58
2.3.4.5 Candidate Grouping 60
2.3.4.6 Candidate Selection 60
2.3.5 Solving for the Transformation Parameters 62
2.3.6 Measuring Image-Model Curve Distances 63
2.3.7 Optimizing the Transformation Parameters 65
2.3.8 Results and Discussion 65
3. Image Database Retrieval Based on Shape Content 73
3.2 The CSS Matching Algorithm 75
3.3 Global Parameters 79
3.4 Performance Evaluation 80
3.5 Results for Original Method 82
3.5.1 CSS without Global Parameters 82
3.5.1.1 CSS without Mirror-Image 83
3.5.2 CSS with Global Parameters 84
3.6 Comparison to Other Methods 86
3.6.1 Fourier Descriptors 86
3.6.2 Moment Invariants 88
3.7 The Problem of Shallow Concavities 90
3.7.1 Height Adjusted Method 92
3.7.1.1 New Maxima 92
3.7.1.2 Shape Segmentation 94
3.7.1.3 A New Global Parameter 95
3.7.2 Average Curvature Method 95
3.7.2.1 Normalization of Average Curvature 96
3.7.2.2 Matching Algorithm 96
3.7.3 Mean-Distance Method 96
3.8 Performance Evaluation and Experimental Results 97
3.8.1 Height Adjusted CSS image 97
3.8.1.1 Without Global Parameter 98
3.8.1.2 With Global Parameter 99
3.8.2 Average Curvature Method 101
3.8.3 Mean-Distance Method 102
3.9 Application to Chrysanthemum Leaf Classification 103
3.9.1 The Problem of Leaf Classification 104
3.9.2 The Problem of Self-Intersection 105
3.9.3 Image Segmentation 105
3.9.4 CSS Image of Self-Crossing Boundaries 108
3.9.5 Recovering Maxima and Minima of CSS Contours 109
3.9.6 Matching CSS minima 111
4. CSS Under Affine Transforms / Non-Rigid Deformations 115
4.2 CSS Image under Affine Transforms 117
4.3 Affine Transforms and Affine Databases 119
4.4 Affine Length 125
4.5 Affine Curvature 126
4.6 Implementation Issues 129
4.6.1 Affine Length 129
4.6.2 Affine Curvature 129
4.7 Experiments and Results 130
4.7.1 Affine Length 130
4.7.2 Affine Curvature 132
4.8 Comparison to Other Methods 132
5. Free-Form 3-D Object Retrieval from Arbitrary Viewpoints 137
5.2 Multi-View 3-D Object Representation and Retrieval 140
5.3 Robust Automatic Selection of Optimal Views 146
5.3.1 Optimal View Selection 147
5.3.2 Combining Optimal Views 149
5.3.3 Recognition Experiment 149
5.3.4 Results 151
5.4 Free-form 3-D Object Retrieval with Occlusion from Arbitrary Viewpoints 154
5.4.1 System Overview 155
5.4.2 Feature Extraction 157
5.4.2.1 Multi-Scale Edge Detection 157
5.4.2.2 Multi-Scale Contour Segmentation 158
5.4.2.3 Segment Features 160
5.4.3 Matching 161
5.4.3.1 Indexing 162
5.4.3.2 Verification through Registration 162
5.4.3.3 Verification through Clustering 165
5.4.4 Experiments on Retrieval with Occlusion 167
6. MPEG-7 Standardisation of the CSS Shape Descriptor 173
6.2.1 Parts of the MPEG-7 Standard 176
6.2.2 MPEG-7 Visual Part 178
6.3 MPEG-7 Shape Descriptors 178
6.4 Contour-Based Shape Descriptor 179
6.4.1 The Contour Shape Descriptor 180
6.4.1.1 Global Parameters 181
6.4.1.2 CSS Peak Parameters 181
6.4.2 Efficient Representation of Descriptor Parameters 183
6.4.3 Matching of the Contour Shape Descriptors 186
6.4.3.1 First Stage of Matching 187
6.4.3.2 Second Stage of Matching 189
6.4.4 Properties of the Contour Shape Descriptor 192
6.5 Region-Based Shape Descriptor 194
6.5.1 The ART Descriptor 195
6.5.2 Similarity Measure 195
6.6 MPEG-7 Performance Testing Methodology and Test-Sets 197
6.6.1 The MPEG-7 Test Database 197
6.6.1.1 Similarity-Based Retrieval 198
6.6.1.2 Rotational and Scaling Invariance 199
6.6.1.3 Robustness to Non-Rigid Motion as well as Other Deformations 200
6.7 Experimental Performance Analysis and MPEG-7 Selection Process 202
6.7.1 Techniques Participating in MPEG-7 Testing 202
6.7.1.1 Wavelet-Based Shape Descriptor 202
6.7.1.2 Polygon-Based Representation 202
6.7.1.3 Fourier-Based Representation 202
6.7.1.4 Multilayer Eigenvector Shape Descriptor 203
6.7.1.5 Zernike Moment-Based Representation 203
6.7.2 Experimental Results 203
6.8 Example Applications of the CSS Shape Descriptor 206
6.8.1 Cartoon Search Engine 206
6.8.2 Object Recognition System 209
7. Robust Image Corner Detection Through Curvature Scale Space 215
7.2 Literature Survey 216
7.3 Canny Edge Detector 217
7.4 Original CSS Corner Detection Method 218
7.4.3 Canny Edge Detection 220
7.4.4 Filling the Gaps and T-junctions 220
7.4.5 Initial Corner Points 220
7.4.6 Tracking 221
7.4.7 Removing False Corners 221
7.5 Original CSS Experimental Results and Discussion 222
7.6 Enhanced CSS Corner Detection Method 225
7.6.1 Using Different Scales of CSS 227
7.6.2 Smoothing the Absolute Curvature Function of Long Contours 229
7.6.3 Tracking 231
7.6.4 Unifying Close Corners 232
7.7 New-CSS Experimental Results and Discussion 232
7.8 Performance Evaluation of Corner Detection Algorithms under Similarity and Affine Transforms 233
7.8.1 Previous Criteria for Performance Measurement 235
7.8.2 Definition of New Criteria 235
7.8.2.1 Consistency 236
7.8.2.2 Accuracy 237
7.8.3 Performance Evaluation, Results and Discussion 238
8. Fast Active Contour Convergence Through CSS Filtering 243
8.2 Literature Survey 246
8.3 Smoothed Active Contour (SAC) 247
9. Multi-Scale Contour Data Compression and Reconstruction Using CSS 255
9.2 Spline Fitting Techniques 256
9.3 Contour Data Reconstruction through CSS and Hermite Curves 257
9.4 Approximation Error and Compression Ratio 258
9.5 Results and Discussion 259
10. Multi-Scale Representations for Free-Form Space Curves 265
10.2 Literature Survey 269
10.3 The Torsion Scale Space Image 269
10.3.1 The Parametric Representation of a Space Curve 270
10.3.2 Computation of Torsion 270
10.3.3 Computing Torsion at Varying Levels of Detail 270
10.3.4 A Multi-Scale Representation for Space Curves 271
10.4 The Renormalized Torsion Scale Space Image 273
10.5 The Resampled Torsion Scale Space Image 276
10.6 Evolution Properties of Space Curves 280
10.7 Space Curve Matching through the TSS Image 283
10.7.1 Torsion Scale Space Matching 284
10.7.2 Solving for the Transformation Parameters 286
10.7.3 Measuring Space Curve Distances 289
10.7.4 Optimizing the Transformation Parameters 289
10.8 Experiments, Discussion and Evaluation 290
11. Multi-Scale Representations for Free-Form 3-D Surfaces 297
11.3 Semigeodesic and Geodesic Polar Parametrisation on a 3-D surface 301
11.3.1 Geodesic Lines 302
11.3.2 Semigeodesic Coordinates 302
11.3.3 Geodesic Polar Coordinates 303
11.3.4 Gaussian Smoothing of a 3-D Surface 303
11.3.5 Multi-Scale Description of a 3-D Surface 304
11.3.6 Implementation on a 3-D Triangular Mesh 304
11.3.6.1 Implementation of Semigeodesic Coordinates 305
11.3.6.2 Semigeodesic Coordinates on Open Surfaces 306
11.3.6.3 Implementation of Geodesic Polar Coordinates 307
11.4 Evolution Properties of 3-D surfaces 307
11.5 Curvature Estimation 308
11.5.1 Curvature Zero-Crossing Contours 310
11.5.2 Local Curvature Maxima 310
11.5.3 Torsion Maxima on Zero-Crossing Contours 311
11.6.1 Diffusion 311
11.6.2 Curvature Estimation 316
11.7 Estimation of Error in Curvature Computation 323
11.8 Robust Free-Form 3-D Object Recognition 335
11.8.1 The Geometric Hashing Algorithm 344
11.8.2 Global Verification 346
11.8.3 Recognition System Results and Discussion 348.