The book provides an excellent introduction to many areas of modern computer vision and machine learning. Although the book is primarily concerned with high-level vision problems such as face recognition, it provides a valuable compendium on a variety of algorithms and statistical techniques which have come into vogue over the last few years.
Complex topics such as statistical learning theory are explained very clearly and concisely with just the right amount of mathematics. The mathematical descriptions are of a kind familiar to anyone with a good science or engineering background, i.e. they are designed to provide insight and formal structure rather than endless proofs of pointless theorems.
I particularly liked the way in which the authors clearly formulate and motivate the key problems before presenting and comparing different solutions. Although somewhat expensive, this book is an excellent introduction and reference for anyone seriously interested in the subject.
Author(s): Shaogang Gong, Stephen J. McKenna
Series: Image Processing
Publisher: World Scientific Publishing Company
Year: 2000
Language: English
Pages: 366
Contents......Page 8
Preface......Page 16
PART I BACKGROUND......Page 22
1.1 The Visual Face......Page 24
1.2 The Changing Face......Page 25
1.3 Computing Faces......Page 27
1.4 Biological Perspectives......Page 32
1.5 The Approach......Page 33
2 Perception and Representation......Page 36
2.1 A Distal Object......Page 37
2.2 Representation by 3D Reconstruction......Page 39
2.3 Two-dimensional View-based Representation......Page 40
2.4 Image Template-based Representation......Page 42
2.5 The Correspondence Problem and Alignment......Page 43
2.6 Biological Perspectives......Page 47
2.7 Discussion......Page 50
3 Learning under Uncertainty......Page 52
3.1 Statistical Learning......Page 53
3.2 Learning as Function Approximation......Page 54
3.3 Bayesian Inference and MAP Classification......Page 57
3.4.1 Parametric Models......Page 58
3.4.2 Non-parametric Models......Page 60
3.4.3 Semi-parametric Models......Page 61
3.5 Unsupervised Learning without Density Estimation......Page 62
3.5.2 Clustering......Page 64
3.6 Linear Classification and Regression......Page 65
3.6.2 Linear Support Vector Machines......Page 66
3.7.1 Multi-layer Networks......Page 69
3.7.2 Support Vector Machines......Page 71
3.8 Adaptation......Page 72
3.9 Biological Perspectives......Page 74
3.10 Discussion......Page 75
PART II FROM SENSORY TO MEANINGFUL PERCEPTION......Page 78
4 Selective Attention: Where to Look......Page 80
4.1 Pre-attentive Visual Cues from Motion......Page 81
4.1.1 Measuring Temporal Change......Page 82
4.1.2 Motion Estimation......Page 83
4.2 Learning Object-based Colour Cues......Page 86
4.2.1 Colour Spaces......Page 87
4.2.2 Colour Density Models......Page 89
4.3 Perceptual Grouping for Selective Attention......Page 92
4.4 Data Fusion for Perceptual Grouping......Page 94
4.5 Temporal Matching and Tracking......Page 97
4.6 Biological Perspectives......Page 98
4.7 Discussion......Page 99
5 A Face Model: What to Look For......Page 102
5.1.1 Feature-based Models......Page 103
5.1.2 Holistic Models......Page 104
5.1.3 The Face Class......Page 106
5.2 Modelling the Face Class......Page 107
5.2.1 Principal Components Analysis for a Face Model......Page 108
5.2.2 Density Estimation in Local PCA Spaces......Page 109
5.3 Modelling a Near-face Class......Page 110
5.4 Learning a Decision Boundary......Page 111
5.4.1 Face Detection in Dynamic Scenes......Page 112
5.4.2 Normalisation......Page 114
5.4.3 Face Detection using Multi-layer Perceptrons......Page 116
5.4.4 Face Detection using Support Vector Machines......Page 118
5.5 Perceptual Search......Page 120
5.6 Biological Perspectives......Page 122
5.7 Discussion......Page 123
6 Understanding Pose......Page 124
6.2 The Face Space across Views: Pose Manifolds......Page 126
6.3 The Effect of Gabor Wavelet Filters on Pose Manifolds......Page 132
6.4 Template Matching as Affine Transformation......Page 134
6.5 Similarities to Prototypes across Views......Page 139
6.6 Learning View-based Support Vector Machines......Page 142
6.8 Discussion......Page 144
7 Prediction and Adaptation......Page 146
7.1 Temporal Observations......Page 149
7.2 Propagating First-order Markov Processes......Page 150
7.3 Kalman Filters......Page 152
7.4.1 Learning Priors using HMMs and EM......Page 153
7.4.2 Observation Augmented Density Propagation......Page 154
7.5 Tracking Attended Regions......Page 155
7.6 Adaptive Colour Models......Page 157
7.7 Selective Adaptation......Page 161
7.8 Tracking Faces......Page 166
7.9 Pose Tracking......Page 168
7.9.1 Person-specific Pose Tracking......Page 170
7.9.2 Person-independent Pose Tracking......Page 171
7.10 Biological Perspectives......Page 177
7.11 Discussion......Page 178
PART III MODELS OF IDENTITY......Page 182
8.1 Identification Tasks......Page 184
8.2 Nearest-neighbour Template Matching......Page 186
8.3 Representing Knowledge of Facial Appearance......Page 187
8.4.1 Low Dimensionality: Principal Components Analysis......Page 189
8.4.2 Separability: Linear Discriminant Analysis......Page 194
8.5 Statistical Knowledge of Identity......Page 197
8.5.1 Identification Tasks Revisited......Page 198
8.5.2 Class-conditional Densities for Modelling Identity......Page 200
8.6 Structural Knowledge: The Role of Correspondence......Page 201
8.6.1 Beyond Alignment: Correspondence at a Single View......Page 202
8.6.2 Combining Statistical and Structural Models......Page 203
8.7 Biological Perspectives......Page 204
8.8 Discussion......Page 206
9 Multi-View Identification......Page 208
9.1 View-based Models......Page 210
9.3 View Correspondence in Identification......Page 212
9.3.1 Learning Linear Shape Models......Page 213
9.3.2 Nonlinear Shape Models......Page 216
9.4.1 Identity by Linear Combination......Page 220
9.4.2 Identity by Similarity to Prototype Views......Page 222
9.5 Generalisation from Multiple Views......Page 224
9.6 Biological Perspectives......Page 226
9.7 Discussion......Page 228
10 Identifying Moving Faces......Page 230
10.1.2 The Effect of Temporal Order of Pose on Learning......Page 231
10.1.3 The Effect of Motion on Familiar Face Identification......Page 232
10.2.1 Atemporal Representation......Page 234
10.2.2 Spatio-temporal Signatures......Page 235
10.3 Identification using Holistic Temporal Trajectories......Page 236
10.4 Identification by Continuous View Transformation......Page 239
10.5 An Experimental System......Page 240
10.6 Discussion......Page 246
PART IV PERCEPTION IN CONTEXT......Page 248
11 Perceptual Integration......Page 250
11.1 Sensory and Model-based Vision......Page 251
11.2 Perceptual Fusion......Page 252
11.3 Perceptual Inference......Page 257
11.3.1 Inference using Hidden Markov Models......Page 261
11.3.2 Closed-loop Perceptual Control......Page 262
11.4.1 Visual Attention and Grouping......Page 263
11.4.2 Face Detection, Tracking and Identification......Page 266
11.5 Biological Perspectives......Page 268
11.6 Discussion......Page 271
12 Beyond Faces......Page 274
12.1 Multi-modal Identification......Page 275
12.2 Visually Mediated Interaction......Page 276
12.3 Visual Surveillance and Monitoring......Page 279
12.4 Immersive Virtual Reality......Page 281
12.5 Visual Database Screening......Page 282
PART V APPENDICES......Page 286
A.1 Database Acquisition and Design......Page 288
A.1.2 Extrinsic Experimental Variables......Page 289
A.2.1 Using Artificial Markers......Page 290
A.2.2 Using Sensors and a Calibrated Camera......Page 292
A.3 Benchmarking......Page 294
A.4 Commercial Databases......Page 296
A.5 Public Domain Face Databases......Page 298
A.6 Discussion......Page 300
B Commercial Systems......Page 302
B.1 System Characterisation......Page 303
B.2 A View on the Industry......Page 304
B.2.1 Visionics Corporation......Page 305
B.2.2 Miros Inc.......Page 307
B.2.3 VisionSpheres Technologies......Page 308
B.2.4 Eigenface-based Systems: Viisage Technologies, Intelligent Verification Systems and Facia Reco......Page 310
B.2.5 Systems based on Facial Feature Matching: Plettac Electronic Security GmbH, ZN Bochum GmbH and Eyematic Interfaces Inc.......Page 312
B.3 Discussion......Page 316
C.1 Principal Components Analysis......Page 318
C.2 Linear Discriminant Analysis......Page 319
C.3.1 Expectation-maximisation......Page 321
C.3.2 Automatic Model Order Selection......Page 322
C.3.3 Adaptive EM for Non-stationary Distributions......Page 323
C.4.1 Zero-order Prediction......Page 325
C.4.2 First-order Prediction......Page 326
C.5 Bayesian Belief Networks......Page 327
C.6 Hidden Markov Models......Page 330
C.7 Gabor Wavelets......Page 333
Bibliography......Page 336
Index......Page 360
