Progress in computer vision and image analysis /

Preface 6
CONTENTS 8
1. An appearance-based method for parametric video registration X. Orriols, L. Barcel麓o and X. Binefa 12
1.1. Introduction 12
1.2. Appearance Based Framework for Multi-Frame Registration 14
1.2.1. Appearance Representation Model 15
1.2.2. Polynomial Surface Model 16
1.2.3. The Algorithm 17
1.3. Experimental Results 18
1.3.1. Selecting a Reference Frame. Consequences in the Registration 19
1.3.2. Analyzing the Complexity in the Polynomial Model. Towards 3D Affine Reconstruction 20
1.4. Summary and Conclusions 25
Acknowledgments 25
References 26
2. An interactive algorithm for image smoothing and segmentation M. C. de Andrade 28
1. Introduction 28
2. The interactive image smoothing and segmentation algorithm - ISS 33
2.1. Edge preserving smoothing under controlled curvature motion 38
2.1.1. Stopping criteria for curvature based denoising 38
2.1.2. Effect of denoising on the ISS 41
2.2. The interactive region growing and merging step 43
2.3. The ISS algorithm steps 45
3. Applications 46
4. Conclusions and Outlook 55
Acknowledgments 56
Appendix A. ISS Pseudo-code 57
Appendix B. ISS Execution time for known test-images 58
References 59
3. Relevance of multifractal textures in static images A. Turiel 62
3.1. Introduction 62
3.2. Multifractal framework 63
3.3. Multifractal decomposition 66
3.4. Reconstructing from edges 70
3.5. Relevance of the fractal manifolds 71
3.6. Conclusions 72
Acknowledgements 77
References 77
4. Potential fields as an external force and algorithmic improvements in deformable models A. Caro et al. 80
4.1. Introduction 81
4.1.1. Overview on Active Contours 81
4.1.2. Scope and purpose of the research 83
4.2. Algorithm Design 84
4.2.1. Standard Deformable Models 84
4.2.2. The new approach for Deformable Models 86
4.2.3. A practical application: Deformable Models on Iberian ham MRI 89
4.3. Practical Results and their Discussion 91
4.4. Conclusions 92
Acknowledgements 94
References 94
5. Optimization of weights in a multiple classifier handwritten word recognition system using a genetic algorithm S. Gunter and H. Bunke 98
5.1. Introduction 98
5.2. Handwritten word recognizer 101
5.2.1. Preprocessing 102
5.2.2. Feature extraction 102
5.2.3. Hidden Markov models 103
5.3. Ensemble creation methods 106
5.3.1. Issues in ensemble creation 106
5.3.2. Bagging 107
5.3.3. AdaBoost 107
5.3.4. Random subspace method 108
5.3.5. Architecture variation 108
5.4. Combination schemes 109
5.4.1. Maximum score rule 109
5.4.2. Performance weighted voting 109
5.4.3. Weighted voting using weights calculated by a genetic algorithm 110
5.4.4. Voting with ties handling 110
5.5. Genetic algorithm for the calculation of the weights used by weighted voting 111
5.5.1. Chromosome representation and fitness 111
5.5.2. Initialization and termination 111
5.5.3. Crossover operator 112
5.5.4. Mutation operator 112
5.5.5. Generation of a new population 112
5.6. Experiments 112
5.7. Conclusions 115
Acknowledgments 116
Appendix A. HandwrittenWord Samples 117
References 118
6. Dempster-Shafer\u2019s basic probability assignment based on fuzzy membership functions A. O. Boudraa et al. 122
6.1. Introduction 122
6.2. Dempster-Shafer theory 123
6.3. Fuzzy approach 125
6.4. Basic probability assignment 126
6.5. Results 130
6.6. Conclusion 131
References 133
7. Automatic instrument localization in laparoscopic surgery J. Climent and P. Mars 134
7.1. Introduction 134
7.2. System Description 135
7.2.1. Filtering stage 136
7.2.2. Edge orientation extraction 136
7.2.3. Hough transform computation 137
7.2.4. Segment extraction 139
7.2.5. Heuristic filter 139
7.2.6. Position prediction 139
7.2.7. Target selection 141
7.3. Results 141
7.4. Conclusion 143
Acknowledgements 145
References 145
8. A fast fractal image compression method based on entropy M. Hassaballah, M. M. Makky and Y. B. Mahdy 148
1. Introduction 148
2. Fractal Image Coding 151
2.1. Principle of Fractal Coding 151
2.2. Baseline Fractal Image Coding Algorithm 153
3. The Proposed Method 154
3.1. Entropy 154
3.2. The Entropy Based Encoded Algorithm 154
4. Experimental Results 157
5. Conclusions 161
References 162
9. Robustness of a blind image watermark detector designed by orthogonal projection C. Jin and J. Peng 164
1. Introduction 164
2. The Design Method of the Watermark Detector 166
3. Experiment Results and Discussion 170
3.1. Performance Test of Two kinds of Methods 170
3.2. Test of Anti-Noise Attack 171
3.3. Test of Anti-Rotation Attack 173
3.4. Test of Anti-Translation Attack 174
3.5. Test of Anti-Other Attack 175
4. Conclusion 176
References 176
10. Self-supervised adaptation for on-line script text recognition L. Prevost and L. Oudot 178
10.1. Introduction 178
10.2. Literature review 179
10.3. Writer independent baseline system 180
10.4. Writer adaptation strategies 183
10.4.1. Supervised adaptation 184
10.4.2. Self-supervised adaptation 185
10.4.2.1. Systematical activation (SA) 186
10.4.2.2. Conditional activation (CA) 186
10.4.2.3. Dynamic management (DM) 187
10.5. Supervised / self-supervised combination 189
10.6. Conclusions & Future works 190
References 191
11. Combining model-based and discriminative approaches in a modular two-stage classification system: Application to isolated handwritten digit recognition J. Milgram, R. Sabourin and M. Cheriet 192
1. Introduction 193
2. Model-based approach 195
2.1. Characterization of the pattern recognition problem 196
2.2. Modeling data with hyperplanes 197
2.3. Estimate posterior probability 199
3. Combination with discriminative approach 200
3.1. Conflict detection 200
3.2. Use of Support Vector Classifiers 200
3.3. Re-estimate posterior probabilities 202
4. Experimental results 203
4.1. Model-based approach 205
4.2. Support Vector Classifiers 208
4.3. Two-stage classification system 209
5. Conclusions and perspectives 213
References 214
12. Learning model structure from data: An application to on-line handwriting H. Binsztok and T. Artieres 218
12.1. Introduction 218
12.2. Building an initial HMM from training data 220
12.2.1. Building a left-right HMM from a training sequence 220
12.2.1.1. HMM structure 221
12.2.1.2. Parameters 221
12.2.2. Building the initial global HMM 223
12.3. Iterative simplification algorithm 225
12.4. Using the approach for clustering and for classification 225
12.5. Experimental databases 226
12.5.1. Artificial data 226
12.5.2. On-line handwritten signals 228
12.6. Probability density function estimation 228
12.6.1. Artificial data 229
12.6.2. Handwritten signals 229
12.7. Clustering experiments 230
12.7.1. Evaluation criteria 231
12.7.2. Benchmark method 231
12.7.3. Experiments on artificial data 232
12.7.4. Experiments on on-line handwritten signals 234
12.8. Classification experiments 236
12.9. Conclusion 237
References 238
13. Simultaneous and causal appearance learning and tracking J. Melenchon, I. Iriondo and L. Meler 242
13.1. Introduction 242
13.2. Incremental SVD with Mean Update 243
13.2.1. Fundamentals 243
13.2.2. Updating SVD 244
13.2.3. Updating SVD and Mean 244
13.2.4. Mean Extraction from a Given SVD 245
13.2.5. Time and Memory Complexity 246
13.3. On-the-Fly Face Training 246
13.3.1. Data Representation 246
13.3.2. Training Process 247
13.3.3. Cost Analysis 248
13.4. Experimental Results 250
13.4.1. On-the-Fly Training Algorithm 250
13.4.2. Incremental SVD and Mean Computation 251
13.4.2.1. Precision comparisons 251
13.4.2.2. Execution time 253
13.4.2.3. Conclusions 253
13.5. Concluding Remarks 254
References 254
14. A comparison framework for walking performances using aSpaces J. Gonzalez et al. 256
14.1. Introduction 256
14.2. RelatedWork 257
14.3. Defining the Training Samples 258
14.2. RelatedWork 257
14.3. Defining the Training Samples 258
14.4. The aWalk aSpace 261
14.5. Parametric Action Representation: the p\u2013action 263
14.6. Human Performance Comparison 265
14.7. Arc length Parameterization of p\u2013actions 266
14.8. Experimental Results 267
14.9. Conclusions and Future Work 269
Acknowledgements 270
References 270
15. Detecting human heads with their orientations A. Sugimoto, M. Kimura and T. Matsuyama 272
15.1. Introduction 272
15.2. Contour model for human-head appearances 274
15.2.1. Human head and its appearances 274
15.2.2. Evaluation of contour model 276
15.3. Inner models for face orientations 279
15.3.1. Facial components 279
15.3.2. Detecting facial components using Gabor-Wavelets 280
15.3.3. Inner models of head appearances with facial components 281
15.4. Algorithm 283
15.5. Experimental evaluation 284
15.5.1. Evaluation on face orientations using a face-image database 284
15.5.2. Evaluation in the real situation 286
15.5.2.1. Human-head detection in the real situation 286
15.5.2.2. Effectiveness of human-head evaluation 288
15.6. Conclusion 290
Acknowledgements 291
References 291
16. Prior knowledge based motion model representation A. D. Sappa et al. 294
16.1. Introduction 294
16.2. PreviousWorks 295
16.3. The Proposed Approach 297
16.3.1. Body Modeling 298
16.3.2. Feature Point Selection and Tracking 300
16.3.2.1. Feature Point Selection 301
16.3.2.2. Feature Point Tracking 301
16.3.3. Motion Model Tuning 301
16.4. Experimental Results 307
16.5. Conclusions and FutureWork 308
Acknowledgements 309
References 309
17. Combining particle filter and population-based metaheuristics for visual articulated motion tracking J. J. Pantrigo et al. 312
17.1. Introduction 313
17.2. Particle Filters 315
17.3. Population-Based Metaheuristics 316
17.3.1. Path Relinking 317
17.3.2. Scatter Search 318
17.4. Particle Filter and Population-Based Metaheuristics Hybrid Algorithms 318
17.4.1. Path Relinking Particle Filter 319
17.4.2. Scatter Search Particle Filter 321
17.4.3. PRPF and SSPF Main Features 322
17.5. Models for Human Pose Estimation 323
17.5.1. Geometrical Model 324
17.5.2. Observation Model and Weighting Function 325
17.5.2.1. System Model 327
17.6. Experimental Results 328
17.7. Conclusion 330
References 330
18. Ear biometrics based on geometrical feature extraction M. Chora麓s 332
1. Introduction 333
2. Ear Biometrics 335
3. Geometrical Method of Feature Extraction 337
3.1. Contour Detection 338
3.2. Normalization 339
3.3. Feature Extraction 340
4. Classification 344
5. Experimental Results and Future Work 345
6. Conclusions 347
References 348
19. Improvement of modal matching image objects in dynamic pedobarography using optimization techniques J. M. R. S. Tavares and L. Ferreira Bastos 350
1. Introduction 350
1.1. Background 351
2. Dynamic Pedobarography 352
3. Object Models 354
3.1. Contour Model 356
3.2. Surface Model 357
3.3. Isobaric Contour Model 360
4. Matching Methodology 362
5. Results 367
5.1. Contour Object Matching 367
5.2. Surface Matching 370
5.3. Isocontour Matching 371
6. Conclusions 374
Acknowledgments 377
References 377
20. Trajectory analysis for sport and video surveillance Y. Lopez de Meneses et al. 380
20.1. Introduction 380
20.2. Point Distribution Models for Trajectories 381
20.2.1. Outlier detection 382
20.2.2. Analysis of temporal information 383
20.3. Experiments 383
20.3.1. Purely-spatial analysis 385
20.3.2. Spatiotemporal analysis 387
20.4. Conclusion 389
20.4.1. Perspectives 389
Acknowledgments 390
References 390
21. Area and volume restoration in elastically deformable solids M. Kelager, A. Fleron and K. Erleben 392
1. Introduction 392
1.1. Background 393
1.2. Motivation 394
1.3. Overview 394
2. Elastically Deformable Solids 395
2.1. Energy of Deformation 395
2.2. Discretization 396
3. Instabilities 398
4. Improvements 399
4.1. Improved Area Restoration 399
4.2. Volume Restoration 401
5. Implosions 402
6. Results 404
7. Conclusion 406
References 410
22. Hand tracking and gesture recognition for human-computer interaction C. Manresa-Yee et al. 412
22.1. Introduction 412
22.2. Hand Segmentation Criteria 414
22.3. Tracking Procedure 416
22.4. Gesture Recognition 417
22.5. System\u2019s Performance Evaluation 420
22.6. Conclusions 421
Acknowledgements 422
References 422
23. A novel approach to sparse histogram image lossless compression using JPEG2000 M. Aguzzi and M. G. Albanesi 424
23.1. Introduction 424
23.2. JPEG2000 Overview 425
23.2.1. Preprocessing 426
23.2.2. Wavelet transform 426
23.2.3. Coefficient coding 427
23.2.3.1. Significance pass 428
23.2.3.2. Refinement pass 429
23.2.3.3. Cleanup pass 429
23.2.4. MQ-coder 430
23.2.5. Tier-2 coder 430
23.3. Toward the Proposed Algorithms 430
23.3.1. The sparsity index: First term computation 431
23.3.2. The sparsity index: Second term computation 431
23.4. The Two Proposals 435
23.4.1. First algorithm: Stripe lengthening 435
23.4.1.1. Single combination experiments 436
23.4.1.2. Overall graphs 440
23.4.1.3. Comparison to other compression algorithms 442
23.4.2. Second algorithm: SuperRLC 446
23.4.2.1. Theoretical considerations 446
23.4.2.2. Comparisons with other images and the sparsity index 449
23.4.2.3. Overall comparison to JPEG2000 450
23.4.2.4. Technical details 453
23.5. Conclusions 454
References 454
24. Genetic programming for object detection: A two-phase approach with an improved fitness function M. Zhang, U. Bhowan and B. Ny 458
24.1. Introduction 458
24.2. Background 460
24.2.1. Object Detection/Recognition and Related Methods 460
24.2.1.1. Performance Evaluation 461
24.2.2. GP Main Characteristics: GP vs GAs 461
24.2.3. GP Related Work to Object Detection 461
24.3. The Approach 462
24.3.1. Overview of the Approach 462
24.3.2. Terminal Set and Function Set 464
24.3.3. Object Classification Strategy 465
24.3.4. Fitness Functions 466
24.3.4.1. The new fitness function 468
24.3.5. Parameters and Termination Criteria 469
24.4. Image Data Sets 470
24.5. Results and Discussion 471
24.5.1. Object Detection Results 471
24.5.2. Training Time and Program Size 472
24.5.3. Comprehensibility of Genetic Programs 473
24.6. Conclusions 474
Acknowledgements 475
References 475
25. Architectural scene reconstruction from single or multiple uncalibrated images H.-Y. Lin, S.-L. Chen and J.-H. Lin 482
25.1. Introduction 482
25.2. Camera Model and Parameter Estimation 484
25.3. Three-Dimensional Model Reconstruction 486
25.3.1. Reconstruction Algorithm 486
25.3.2. Registration and Pose Estimation 487
25.3.3. Model Optimization 488
25.4. Experimental Results 489
25.5. Conclusion and Future Research 490
Acknowledgments 492
References 492
26. Separating rigid motion for continuous shape evolution N. C. Overgaard and J. E. Solem 494
26.1. Introduction 494
26.2. Level Sets, Normal Velocity, and L2-Gradient Descent 495
26.3. Decomposition of Evolutions 497
26.3.1. The Projection onto Translations 497
26.3.2. The Projection onto Rotations 499
26.4. Experiments 500
26.4.1. Continuous Shape Warping 501
26.4.2. Registration of Continuous Shapes 502
26.5. Conclusions 503
References 504
27. A PDE method to segment image linear objects with application to lens distortion removal M. T. El-Melegy and N. H. Al-Ashwal 506
1. Introduction 506
2. Region-Based Segmentation with Level Sets and PDEs 508
3. PDE Method for Line Segmentation 510
4. Multi-Object Segmentation 513
4.1. The Fuzzy C-mean Algorithm 513
4.2. Handling Multiple Objects 514
5. Experimental Results 517
6. Application: Lens Distortion Removal 518
6.1. Camera Distortion Model 522
6.2. Our Approach 522
6.3. Experimental Results 523
7. Conclusions 525
References 526
28. Improved motion segmentation based on shadow detection M. Kampel et al. 530
28.1. Introduction 530
28.2. Colour Spaces 531
28.2.1. Normalised RGB 531
28.2.2. IHLS Space 532
28.2.3. Hue Statistics 533
28.2.4. Saturation-Weighted Hue Statistics 535
28.3. The IHLS Background Model 536
28.4. Metrics forMotion Segmentation 537
28.5. Experiments and Results 539
28.6. Conclusion 542
References 543
29. SnakeCut: An integrated approach based on active contour and GrabCut for automatic foreground object segmentation S. Prakash, R. Abhilash and S. Das 546
29.1. Introduction 546
29.2. Preliminaries 548
29.2.1. Active Contour (Snake)Model 548
29.2.2. GrabCut 549
29.3. Comparison of Active Contour and GrabCut Methods 550
29.4. SnakeCut: Integration of Active Contour and GrabCut 553
29.5. SnakeCut Segmentation Results 557
29.6. Conclusion 564
References 565
30. Intelligent CCTV for mass transport security: Challenges and opportunities for video and face processing C. Sanderson et al. 568
30.1. Introduction 569
30.2. Bag-of-Features Approaches 572
30.2.1. Feature Extraction and Illumination Normalisation 572
30.2.2. Bag-of-Features with Direct Likelihood Evaluation 572
30.2.3. Bag-of-Features with Histogram Matching 573
30.2.4. Speedup via Approximation 574
30.3. Active Appearance Models 574
30.3.1. Face Modelling 575
30.3.2. Pose Estimation 576
30.3.3. Frontal View Synthesis 576
30.3.4. Direct Pose-Robust Features 577
30.4. Evaluation 578
30.5. Discussion 580
Acknowledgements 582
References 582
Author Index 586
Subject Index 588