Object Categorization: Computer and Human Vision Perspectives

Half-title......Page 3
Title......Page 5
Copyright......Page 6
Contents......Page 7
Preface......Page 11
Contributors......Page 15
1.1 Introduction......Page 19
1.2 Avoiding the Abstraction Problem: A Historical Trend......Page 21
1.3 The Abstraction of Shape......Page 28
1.4 The Abstraction of Structure......Page 32
1.5 Segmentation, Grouping, and the Role of Models: Beyond Target Recognition......Page 34
1.6 Expanding Model Scope: Objects to Scenes......Page 36
1.7 Managing Search Complexity: The Case for 3-D Models......Page 38
1.8 Identifying Our Shortcomings: The Need for New Benchmarks......Page 40
1.9 Conclusion......Page 42
Acknowledgments......Page 43
Bibliography......Page 44
2.1 Introduction......Page 56
2.3 Can We Agree on the Overarching Goal?......Page 57
2.5 Establishing Benchmarks for Core Recognition......Page 59
2.6 What Computational Processes Must Be at Work to Accomplish Core Recognition?......Page 61
2.7 What Does a Good Representation to Support Core Recognition Look Like?......Page 64
2.8 How Might the Ventral Visual Stream Build Good Representation?......Page 65
2.9 A Strategy for Understanding Visual Object Recognition......Page 68
Bibliography......Page 69
3.2 What?......Page 73
3.3 Why?......Page 74
3.4 Which?......Page 75
3.5 How Many? How Fast?......Page 76
3.7 The State of the Art......Page 78
3.8 Visual Learning......Page 81
3.9 Towards Scene Understanding......Page 82
3.10 Conclusion......Page 84
Bibliography......Page 85
4.1 Introduction......Page 87
4.2 Seeing vs. "Seeing As"......Page 88
4.3 A Closer Look at "Seeing As"......Page 91
4.4.1 The Role of the Task......Page 92
4.4.2 The Role of Context......Page 94
4.4.3 The Role of Conceptual Knowledge......Page 96
4.5 Some Parallels with Biological Vision......Page 98
4.6 Conclusions......Page 99
Bibliography......Page 100
5.1 Introduction......Page 105
5.2 A Principled Approach to Generic Object Recognition......Page 107
5.3.1 Part-Based Representations......Page 108
5.3.1.1 Volumes vs. Surfaces......Page 109
5.4 Computation of Part-Based Descriptions......Page 110
5.4.2 Generating Descriptions......Page 111
5.4.3 A Hierarchy......Page 112
5.5 Recognition Using Part-Based Descriptions......Page 113
5.5.3.1 Defining Similarities......Page 114
5.5.3.2 Retrieval......Page 115
5.6 TheWay Forward......Page 116
Bibliography......Page 117
6.1 Introduction......Page 120
6.2 The Functional Organization of the Human Ventral Stream......Page 121
6.3 Cue-Invariant Responses in the LOC......Page 123
6.4 Neural Bases of Invariant Object Recognition......Page 126
6.5 Object and Position Information in the LOC......Page 127
6.6 Evidence for Viewpoint Sensitivity Across the LOC......Page 131
6.7 Debates About the Nature of Functional Organization in the Human Ventral Stream......Page 135
6.7.2 Process Maps......Page 136
6.7.4 Topographic Representation......Page 137
6.8.1 fMRI Measurements of the Development of the Ventral Stream......Page 138
6.8.2 Implications of Differential Development of Visual Cortex......Page 140
6.9 Conclusion......Page 141
Bibliography......Page 142
7.1 Recognition Based on Functionality......Page 147
7.2.1 Reasoning about Static 3-D Shapes......Page 149
7.2.2 Functionality Requiring Articulated Shape......Page 150
7.2.4 Robotic Interaction with Objects......Page 151
7.3.1 Functionality-Related Work within AI......Page 152
7.3.2 Functionality-Based Computer Vision......Page 154
7.3.3 Reasoning about Functionality in Robotics......Page 156
7.4 Open Problem Areas......Page 159
7.4.2 Scalability and Efficiency......Page 160
7.4.4 Scalability and Learning......Page 161
Bibliography......Page 162
8.1 Introduction......Page 166
8.2 The Conventional View......Page 167
8.3 The Conventional Evolutionary Argument......Page 168
8.4 Bayes' Circle......Page 169
8.5 The Interface Theory of Perception......Page 171
8.6 User Interfaces in Nature......Page 173
8.7 Interface and World......Page 176
8.8 Future Research on Perceptual Categorization......Page 178
8.9 Interface Games......Page 179
8.10 Conclusion......Page 181
Bibliography......Page 182
9.2 Word Frequencies, Objects, and Scenes......Page 185
9.2.1 Exploiting Correlations between Words and Pictures......Page 186
9.2.2 Scenes and Annotations......Page 188
9.2.3 Driving Recognition with Word-Picture Relations......Page 190
9.3 Syntax, Depiction, Modifiers, and Iconography......Page 191
9.3.1 Depiction......Page 192
9.3.2 Modifiers and Attributes......Page 193
9.3.3 Iconography......Page 194
Bibliography......Page 195
10.1 Introduction......Page 200
10.2 Is Biological Object Vision Explicitly Structural?......Page 202
10.3 Is Computational Vision Secretly Structural?......Page 207
10.4 Reconciling Biological and Computer Vision......Page 210
Bibliography......Page 211
11.1 Introduction......Page 214
11.2.1 Hierarchical Compositionality......Page 216
11.2.2 Statistical, Bottom-up Learning......Page 219
11.3 Related Work......Page 220
11.4 Learning a Compositional Hierarchy of Parts......Page 221
11.4.1 The Compositional Library of Parts......Page 222
11.4.2 Hierarchical Recognition: The Indexing and Matching Scheme......Page 223
11.4.3 Unsupervised Learning of Part Compositions......Page 225
11.4.4 Grouping Part Labels by Similarity and Co-occurrence......Page 226
11.5 Experimental Results......Page 227
11.6 Conclusion......Page 230
Bibliography......Page 231
12.1 Introduction......Page 234
12.2.1 One Mechanism Fits All? The Benchmark Test Case of Faces......Page 237
12.3.1 Theories of Perceptual Categorization......Page 239
12.3.2 Neural Bases of Categorization: Human fMRI Data......Page 240
12.3.3 Neural Bases of Categorization: Monkey Electrophysiology Data......Page 243
12.3.4 Stimulus-Driven Learning at Lower Levels of the Hierarchy and Its Relevance for Recognition in Clutter......Page 246
12.3.5 Activation versus Representation......Page 247
12.4 What to Do if Rapid Recognition Breaks: Tweaking Things on the Fly......Page 248
12.5 More Open Questions......Page 250
12.6 Conclusion......Page 252
Bibliography......Page 253
13.1 Introduction......Page 259
13.2 Representation......Page 260
13.3 Learning and Estimation with the And-Or Graph......Page 263
13.3.2 Learning and Pursuing the Relation Set......Page 264
13.4 Experiments on Learning and Sampling......Page 265
13.5 Inference with the And-Or Graph......Page 268
13.6 Experiments on Object Recognition Using the And-Or Graph......Page 270
Acknowledgments......Page 272
Bibliography......Page 273
14.2 The Neurophysiology of Object Representation in the Inferior Temporal Visual Cortex......Page 275
14.3 Outline of a Feature Hierarchy Model of the Computational Mechanisms in the Visual Cortex for Object and Face Recognition......Page 278
14.4 A Computational Model of Invariant Visual Object and Face Recognition......Page 280
14.5.1 Object-Based Attention and Object Selection in Complex Natural Scenes......Page 283
14.5.2 The Interface from Object Representations to Action......Page 284
14.6 Object Representation and Attention in Natural Scenes: A Computational Account......Page 287
14.7 Learning Invariant Representations of an Object with Multiple Objects in the Scene and with Cluttered Backgrounds......Page 291
14.8 A Biased Competition Model of Object and Spatial Attentional Effects on the Representations in the Visual System......Page 293
14.9 Decision making in Perception......Page 295
14.11 Conclusion......Page 297
Acknowledgments......Page 300
Bibliography......Page 301
15.1 Introduction......Page 306
15.3 Full Object Interpretation......Page 307
15.3.1 Informative Feature Hierarchies......Page 308
15.3.2 Recognizing Objects and Parts......Page 309
15.3.3 Semantic Features......Page 311
15.4.1 The Performance Gap......Page 313
15.4.2 Incremental Model Selection......Page 315
Bibliography......Page 317
16.1.1 Object Discovery from the Developmental Perspective......Page 319
16.2 Prior Work on Object Discovery......Page 321
16.2.1 Learning from Normalized Images......Page 322
16.2.2 Semi-Supervised Models......Page 323
16.3.1 Orienting towards Interesting Regions......Page 325
16.3.2 Grouping Regions into Larger Assemblies......Page 326
16.3.3 Grouping Diverse Object Instances within the Same Category......Page 328
16.4 Synthesizing the Experimental Data into a Theoretical Framework for Object Discovery......Page 329
16.4.2 M2: Grouping Regions into “Proto-objects”......Page 330
16.4.2.1 Color Similarity–Based Grouping......Page 331
16.4.2.2 Collinearity-based Grouping......Page 333
16.4.3 M3: Tracking Proto-Objects and Constructing Temporally Extended Appearance Models......Page 334
16.5 Conclusion......Page 338
Bibliography......Page 339
17.1 Introduction......Page 342
17.1.1.1 Model Paradigm......Page 343
17.1.1.2 Representation Paradigm......Page 344
17.1.1.3 Learning Paradigm......Page 346
17.1.2 Towards Integration Across Paradigms......Page 347
17.2.2 Probabilistic Topic Models of Localized Gradients......Page 348
17.3 Unsupervised Discovery of Visual Categories......Page 350
17.3.1 Google Re-ranking Task......Page 351
17.3.2 Object Class Discovery......Page 352
17.4.2 Sliding-Window Approach to Detection......Page 355
17.5 Experiments......Page 356
17.5.2 Comparison of Methods for Estimation and Evaluation of Approximate Inference......Page 357
17.5.3 Comparison to State-of-the-Art on PASCAL’06 VOC Detection Challenge......Page 358
17.5.4 Comparison to Shape Features on ETH Shape Database......Page 360
17.6 Conclusion......Page 361
Bibliography......Page 362
18.1 Introduction......Page 366
18.2 Category-Specific Topography of the Ventral Visual Pathway......Page 367
18.3 Typical Development of Face-Related Cortex......Page 368
18.3.1 Mechanisms of Functional Brain Development in Fusiform Gyrus......Page 369
18.3.2 Relating Developmental Changes in Face-Processing Behavior and Face-Related Activation......Page 371
18.4 Atypical Development of Category-Specific Topography in Ventral Visual Cortex......Page 374
18.5 Conclusion......Page 379
Bibliography......Page 381
19.1 Introduction......Page 387
19.2.1 Model Representation and Recognition Algorithm......Page 389
19.2.2 Learning......Page 392
19.2.3 Examples......Page 394
19.3 Grouping......Page 395
19.3.1 Pairwise Grouping of Features......Page 396
19.4 Conclusion......Page 399
Bibliography......Page 400
20.1 Introduction......Page 402
20.2 The Role of Orbitofrontal Cortex in the Activation of Top-Down Predictions......Page 404
20.3 Top-Down Facilitation Model of Object Recognition......Page 405
20.4 The Contribution of Low Spatial Frequencies......Page 406
20.5 The Magnocellular Pathway Mediates Top-Down Facilitation......Page 408
20.6 Contextual Predictions for Object Recognition......Page 411
20.7 Conclusion......Page 414
Bibliography......Page 416
21.1 Introduction......Page 419
21.2 Survey of Related Work......Page 421
21.3.1 Pyramid Match Kernels......Page 422
21.3.2 Spatial Matching Scheme......Page 423
21.4.2 Scene Category Recognition......Page 425
21.4.3 Caltech-101......Page 427
21.4.4 Discussion......Page 428
21.5 Applications and Extensions......Page 429
21.6 Conclusion......Page 430
Bibliography......Page 431
22.1 Introduction......Page 434
22.2 Perceptual Decisions and Learning......Page 435
22.3 Categorical Decisions and Learning......Page 440
22.4 Conclusion......Page 441
Bibliography......Page 443
23.1 Introduction......Page 448
23.2 Shock-Graph Representation of Figure-Ground Segregated Shape......Page 451
23.3 Perceptional Grouping: A Case for Intermediate-Level Representation......Page 455
23.4 Bottom-Up Recognition of Shapes Embedded in Figures......Page 459
23.5 Model-Based Segmentation: Top-Down Organization Using the Shock Graph......Page 461
Bibliography......Page 463
24.1 Introduction......Page 469
24.2.1 Encoding Correlation Structures in Tuning Properties......Page 470
24.2.2 Encoding Competitive Constraints in Neuronal Interaction......Page 473
24.2.3 Encoding Spatial Correlations in Neuronal Connectivity......Page 475
24.3 Computational Implications on Cortical Object Representation......Page 479
24.3.1 IT Horizontal Connections for Encoding Temporal Association of Views......Page 480
24.3.2 V4 Horizontal Connections Encoding Spatial Relationships between Parts......Page 484
24.4 Conclusion......Page 487
Bibliography......Page 489
25.1 Medial Representations of Objects......Page 493
25.1.1 Topological Properties of Medial Loci......Page 495
25.1.2 Local Geometry of Medial Loci......Page 496
25.2 Medial Loci in Human Vision......Page 497
25.3.1 Grassfire Flow......Page 500
25.3.2 Average Outward Flux and Object Angle......Page 501
Bibliography......Page 504
26.1 Introduction......Page 506
26.2.1 Exemplar-Based Versus Structural Object Representations......Page 508
26.2.2 Temporal Information......Page 509
26.2.3 Computational Implementation......Page 510
26.3.1 Exchanging Materials......Page 512
26.4 Multisensory Object Processing......Page 514
26.4.1 Visuohaptic Similarity Ratings......Page 515
26.4.2 Computational Issues......Page 517
Bibliography......Page 518
27.1 Introduction......Page 520
27.2 Comparing Images with Pose Variation......Page 523
27.3 Deformations and Articulations......Page 526
27.4 Lighting......Page 529
27.5 Conclusion......Page 531
Bibliography......Page 532
Index......Page 535