Computer vision a modern approach = 计算机视觉 一种现代方法 (英文版·第二版)PDF|Epub|txt|kindle电子书版本网盘下载

Computer vision a modern approach = 计算机视觉 一种现代方法 (英文版·第二版)PDF格式电子书版下载

注意：本站所有压缩包均有解压码： 点击下载压缩包解压工具

Ⅰ IMAGE FORMATION1

1 Geometric Camera Models3

1.1 Image Formation4

1.1.1 Pinhole Perspective4

1.1.2 Weak Perspective6

1.1.3 Cameras with Lenses8

1.1.4 The Human Eye12

1.2 Intrinsic and Extrinsic Parameters14

1.2.1 Rigid Transformations and Homogeneous Coordinates14

1.2.2 Intrinsic Parameters16

1.2.3 Extrinsic Parameters18

1.2.4 Perspective Projection Matrices19

1.2.5 Weak-Perspective Projection Matrices20

1.3 Geometric Camera Calibration22

1.3.1 A Linear Approach to Camera Calibration23

1.3.2 A Nonlinear Approach to Camera Calibration27

1.4 Notes29

2 Light and Shading32

2.1 Modelling Pixel Brightness32

2.1.1 Reflection at Surfaces33

2.1.2 Sources and Their Effects34

2.1.3 The Lambertian＋Specular Model36

2.1.4 Area Sources36

2.2 Inference from Shading37

2.2.1 Radiometric Calibration and High Dynamic Range Images38

2.2.2 The Shape of Specularities40

2.2.3 Inferring Lightness and Illumination43

2.2.4 Photometric Stereo：Shape from Multiple Shaded Images46

2.3 Modelling Interreflection52

2.3.1 The Illumination at a Patch Due to an Area Source52

2.3.2 Radiosity and Exitance54

2.3.3 An Interreflection Model55

2.3.4 Qualitative Properties of Interreflections56

2.4 Shape from One Shaded Image59

2.5 Notes61

3 Color68

3.1 Human Color Perception68

3.1.1 Color Matching68

3.1.2 Color Receptors71

3.2 The Physics of Color73

3.2.1 The Color of Light Sources73

3.2.2 The Color of Surfaces76

3.3 Representing Color77

3.3.1 Linear Color Spaces77

3.3.2 Non-linear Color Spaces83

3.4 A Model of Image Color86

3.4.1 The Diffuse Term88

3.4.2 The Specular Term90

3.5 Inference from Color90

3.5.1 Finding Specularities Using Color90

3.5.2 Shadow Removal Using Color92

3.5.3 Color Constancy：Surface Color from Image Color95

3.6 Notes99

Ⅱ EARLY VISION：JUST ONE IMAGE105

4 Linear Filters107

4.1 Linear Filters and Convolution107

4.1.1 Convolution107

4.2 Shift Invariant Linear Systems112

4.2.1 Discrete Convolution113

4.2.2 Continuous Convolution115

4.2.3 Edge Effects in Discrete Convolutions118

4.3 Spatial Frequency and Fourier Transforms118

4.3.1 Fourier Transforms119

4.4 Sampling and Aliasing121

4.4.1 Sampling122

4.4.2 Aliasing125

4.4.3 Smoothing and Resampling126

4.5 Filters as Templates131

4.5.1 Convolution as a Dot Product131

4.5.2 Changing Basis132

4.6 Technique：Normalized Correlation and Finding Patterns132

4.6.1 Controlling the Television by Finding Hands by Normalized Correlation133

4.7 Technique：Scale and Image Pyramids134

4.7.1 The Gaussian Pyramid135

4.7.2 Applications of Scaled Representations136

4.8 Notes137

5 Local Image Features141

5.1 Computing the Image Gradient141

5.1.1 Derivative of Gaussian Filters142

5.2 Representing the Image Gradient144

5.2.1 Gradient-Based Edge Detectors145

5.2.2 Orientations147

5.3 Finding Corners and Building Neighborhoods148

5.3.1 Finding Corners149

5.3.2 Using Scale and Orientation to Build a Neighborhood151

5.4 Describing Neighborhoods with SIFT and HOG Features155

5.4.1 SIFT Features157

5.4.2 HOG Features159

5.5 Computing Local Features in Practice160

5.6 Notes160

6 Texture164

6.1 Local Texture Representations Using Filters166

6.1.1 Spots and Bars167

6.1.2 From Filter Outputs to Texture Representation168

6.1.3 Local Texture Representations in Practice170

6.2 Pooled Texture Representations by Discovering Textons171

6.2.1 Vector Quantization and Textons172

6.2.2 K-means Clustering for Vector Quantization172

6.3 Synthesizing Textures and Filling Holes in Images176

6.3.1 Synthesis by Sampling Local Models176

6.3.2 Filling in Holes in Images179

6.4 Image Denoising182

6.4.1 Non-local Means183

6.4.2 Block Matching 3D （BM3D）183

6.4.3 Learned Sparse Coding184

6.4.4 Results186

6.5 Shape from Texture187

6.5.1 Shape from Texture for Planes187

6.5.2 Shape from Texture for Curved Surfaces190

6.6 Notes191

Ⅲ EARLY VISION：MULTIPLE IMAGES195

7 Stereopsis197

7.1 Binocular Camera Geometry and the Epipolar Constraint198

7.1.1 Epipolar Geometry198

7.1.2 The Essential Matrix200

7.1.3 The Fundamental Matrix201

7.2 Binocular Reconstruction201

7.2.1 Image Rectification202

7.3 Human Stereopsis203

7.4 Local Methods for Binocular Fusion205

7.4.1 Correlation205

7.4.2 Multi-Scale Edge Matching207

7.5 Global Methods for Binocular Fusion210

7.5.1 Ordering Constraints and Dynamic Programming210

7.5.2 Smoothness and Graphs211

7.6 Using More Cameras214

7.7 Application：Robot Navigation215

7.8 Notes216

8 Structure from Motion221

8.1 Internally Calibrated Perspective Cameras221

8.1.1 Natural Ambiguity of the Problem223

8.1.2 Euclidean Structure and Motion from Two Images224

8.1.3 Euclidean Structure and Motion from Multiple Images228

8.2 Uncalibrated Weak-Perspective Cameras230

8.2.1 Natural Ambiguity of the Problem231

8.2.2 Affine Structure and Motion from Two Images233

8.2.3 Affine Structure and Motion from Multiple Images237

8.2.4 From Affine to Euclidean Shape238

8.3 Uncalibrated Perspective Cameras240

8.3.1 Natural Ambiguity of the Problem241

8.3.2 Projective Structure and Motion from Two Images242

8.3.3 Projective Structure and Motion from Multiple Images244

8.3.4 From Projective to Euclidean Shape246

8.4 Notes248

Ⅳ MID-LEVEL VISION253

9 Segmentation by Clustering255

9.1 Human Vision：Grouping and Gestalt256

9.2 Important Applications261

9.2.1 Background Subtraction261

9.2.2 Shot Boundary Detection264

9.2.3 Interactive Segmentation265

9.2.4 Forming Image Regions266

9.3 Image Segmentation by Clustering Pixels268

9.3.1 Basic Clustering Methods269

9.3.2 The Watershed Algorithm271

9.3.3 Segmentation Using K-means272

9.3.4 Mean Shift：Finding Local Modes in Data273

9.3.5 Clustering and Segmentation with Mean Shift275

9.4 Segmentation，Clustering，and Graphs277

9.4.1 Terminology and Facts for Graphs277

9.4.2 Agglomerative Clustering with a Graph279

9.4.3 Divisive Clustering with a Graph281

9.4.4 Normalized Cuts284

9.5 Image Segmentation in Practice285

9.5.1 Evaluating Segmenters286

9.6 Notes287

10 Grouping and Model Fitting290

10.1 The Hough Transform290

10.1.1 Fitting Lines with the Hough Transform290

10.1.2 Using the Hough Transform292

10.2 Fitting Lines and Planes293

10.2.1 Fitting a Single Line294

10.2.2 Fitting Planes295

10.2.3 Fitting Multiple Lines296

10.3 Fitting Curved Structures297

10.4 Robustness299

10.4.1 M-Estimators300

10.4.2 RANSAC：Searching for Good Points302

10.5 Fitting Using Probabilistic Models306

10.5.1 Missing Data Problems307

10.5.2 Mixture Models and Hidden Variables309

10.5.3 The EM Algorithm for Mixture Models310

10.5.4 Difficulties with the EM Algorithm312

10.6 Motion Segmentation by Parameter Estimation313

10.6.1 Optical Flow and Motion315

10.6.2 Flow Models316

10.6.3 Motion Segmentation with Layers317

10.7 Model Selection：Which Model Is the Best Fit？319

10.7.1 Model Selection Using Cross-Validation322

10.8 Notes322

11 Tracking326

11.1 Simple Tracking Strategies327

11.1.1 Tracking by Detection327

11.1.2 Tracking Translations by Matching330

11.1.3 Using Affine Transformations to Confirm a Match332

11.2 Tracking Using Matching334

11.2.1 Matching Summary Representations335

11.2.2 Tracking Using Flow337

11.3 Tracking Linear Dynamical Models with Kalman Filters339

11.3.1 Linear Measurements and Linear Dynamics340

11.3.2 The Kalman Filter344

11.3.3 Forward-backward Smoothing345

11.4 Data Association349

11.4.1 Linking Kalman Filters with Detection Methods349

11.4.2 Key Methods of Data Association350

11.5 Particle Filtering350

11.5.1 Sampled Representations of Probability Distributions351

11.5.2 The Simplest Particle Filter355

11.5.3 The Tracking Algorithm356

11.5.4 A Workable Particle Filter358

11.5.5 Practical Issues in Particle Filters360

11.6 Notes362

Ⅴ HIGH-LEVEL VISION365

12 Registration367

12.1 Registering Rigid Objects368

12.1.1 Iterated Closest Points368

12.1.2 Searching for Transformations via Correspondences369

12.1.3 Application：Building Image Mosaics370

12.2 Model-based Vision：Registering Rigid Objects with Projection375

12.2.1 Verification：Comparing Transformed and Rendered Source to Target377

12.3 Registering Deformable Objects378

12.3.1 Deforming Texture with Active Appearance Models378

12.3.2 Active Appearance Models in Practice381

12.3.3 Application：Registration in Medical Imaging Systems383

12.4 Notes388

13 Smooth Surfaces and Their Outlines391

13.1 Elements of Differential Geometry393

13.1.1 Curves393

13.1.2 Surfaces397

13.2 Contour Geometry402

13.2.1 The Occluding Contour and the Image Contour402

13.2.2 The Cusps and Inflections of the Image Contour403

13.2.3 Koenderink’s Theorem404

13.3 Visual Events：More Differential Geometry407

13.3.1 The Geometry of the Gauss Map407

13.3.2 Asymptotic Curves409

13.3.3 The Asymptotic Spherical Map410

13.3.4 Local Visual Events412

13.3.5 The Bitangent Ray Manifold413

13.3.6 Multilocal Visual Events414

13.3.7 The Aspect Graph416

13.4 Notes417

14 Range Data422

14.1 Active Range Sensors422

14.2 Range Data Segmentation424

14.2.1 Elements of Analytical Differential Geometry424

14.2.2 Finding Step and Roof Edges in Range Images426

14.2.3 Segmenting Range Images into Planar Regions431

14.3 Range Image Registration and Model Acquisition432

14.3.1 Quaternions433

14.3.2 Registering Range Images434

14.3.3 Fusing Multiple Range Images436

14.4 Object Recognition438

14.4.1 Matching Using Interpretation Trees438

14.4.2 Matching Free-Form Surfaces Using Spin Images441

14.5 Kinect446

14.5.1 Features447

14.5.2 Technique：Decision Trees and Random Forests448

14.5.3 Labeling Pixels450

14.5.4 Computing Joint Positions453

14.6 Notes453

15 Learning to Classify457

15.1 Classification，Error，and Loss457

15.1.1 Using Loss to Determine Decisions457

15.1.2 Training Error，Test Error，and Overfitting459

15.1.3 Regularization460

15.1.4 Error Rate and Cross-Validation463

15.1.5 Receiver Operating Curves465

15.2 Major Classification Strategies467

15.2.1 Example：Mahalanobis Distance467

15.2.2 Example：Class-Conditional Histograms and Naive Bayes468

15.2.3 Example：Classification Using Nearest Neighbors469

15.2.4 Example：The Linear Support Vector Machine470

15.2.5 Example：Kernel Machines473

15.2.6 Example：Boosting and Adaboost475

15.3 Practical Methods for Building Classifiers475

15.3.1 Manipulating Training Data to Improve Performance477

15.3.2 Building Multi-Class Classifiers Out of Binary Classifiers479

15.3.3 Solving for SVMS and Kernel Machines480

15.4 Notes481

16 Classifying Images482

16.1 Building Good Image Features482

16.1.1 Example Applications482

16.1.2 Encoding Layout with GIST Features485

16.1.3 Summarizing Images with Visual Words487

16.1.4 The Spatial Pyramid Kernel489

16.1.5 Dimension Reduction with Principal Components493

16.1.6 Dimension Reduction with Canonical Variates494

16.1.7 Example Application：Identifying Explicit Images498

16.1.8 Example Application：Classifying Materials502

16.1.9 Example Application：Classifying Scenes502

16.2 Classifying Images of Single Objects504

16.2.1 Image Classification Strategies505

16.2.2 Evaluating Image Classification Systems505

16.2.3 Fixed Sets of Classes508

16.2.4 Large Numbers of Classes509

16.2.5 Flowers，Leaves，and Birds：Some Specialized Problems511

16.3 Image Classification in Practice512

16.3.1 Codes for Image Features513

16.3.2 Image Classification Datasets513

16.3.3 Dataset Bias515

16.3.4 Crowdsourcing Dataset Collection515

16.4 Notes517

17 Detecting Objects in Images519

17.1 The Sliding Window Method519

17.1.1 Face Detection520

17.1.2 Detecting Humans525

17.1.3 Detecting Boundaries527

17.2 Detecting Deformable Objects530

17.3 The State of the Art of Object Detection535

17.3.1 Datasets and Resources538

17.4 Notes539

18 Topics in Object Recognition540

18.1 What Should Object Recognition Do？540

18.1.1 What Should an Object Recognition System Do？540

18.1.2 Current Strategies for Object Recognition542

18.1.3 What Is Categorization？542

18.1.4 Selection：What Should Be Described？544

18.2 Feature Questions544

18.2.1 Improving Current Image Features544

18.2.2 Other Kinds of Image Feature546

18.3 Geometric Questions547

18.4 Semantic Questions549

18.4.1 Attributes and the Unfamiliar550

18.4.2 Parts，Poselets and Consistency551

18.4.3 Chunks of Meaning554

Ⅵ APPLICATIONS AND TOPICS557

19 Image-Based Modeling and Rendering559

19.1 Visual Hulls559

19.1.1 Main Elements of the Visual Hull Model561

19.1.2 Tracing Intersection Curves563

19.1.3 Clipping Intersection Curves566

19.1.4 Triangulating Cone Strips567

19.1.5 Results568

19.1.6 Going Further：Carved Visual Hulls572

19.2 Patch-Based Multi-View Stereopsis573

19.2.1 Main Elements of the PMVS Model575

19.2.2 Initial Feature Matching578

19.2.3 Expansion579

19.2.4 Filtering580

19.2.5 Results581

19.3 The Light Field584

19.4 Notes587

20 Looking at People590

20.1 HMM’s，Dynamic Programming，and Tree-Structured Models590

20.1.1 Hidden Markov Models590

20.1.2 Inference for an HMM592

20.1.3 Fitting an HMM with EM597

20.1.4 Tree-Structured Energy Models600

20.2 Parsing People in Images602

20.2.1 Parsing with Pictorial Structure Models602

20.2.2 Estimating the Appearance of Clothing604

20.3 Tracking People606

20.3.1 Why Human Tracking Is Hard606

20.3.2 Kinematic Tracking by Appearance608

20.3.3 Kinematic Human Tracking Using Templates609

20.4 3D from 2D：Lifting611

20.4.1 Reconstruction in an Orthographic View611

20.4.2Exploiting Appearance for Unambiguous Reconstructions613

20.4.3Exploiting Motion for Unambiguous Reconstructions615

20.5Activity Recognition617

20.5.1 Background：Human Motion Data617

20.5.2 Body Configuration and Activity Recognition621

20.5.3 Recognizing Human Activities with Appearance Features622

20.5.4 Recognizing Human Activities with Compositional Models624

20.6Resources624

20.7Notes626

21 Image Search and Retrieval627

21.1 The Application Context627

21.1.1 Applications628

21.1.2 User Needs629

21.1.3 Types of Image Query630

21.1.4 What Users Do with Image Collections631

21.2 Basic Technologies from Information Retrieval632

21.2.1 Word Counts632

21.2.2 Smoothing Word Counts633

21.2.3 Approximate Nearest Neighbors and Hashing634

21.2.4 Ranking Documents638

21.3 Images as Documents639

21.3.1 Matching Without Quantization640

21.3.2 Ranking Image Search Results641

21.3.3 Browsing and Layout643

21.3.4 Laying Out Images for Browsing644

21.4 Predicting Annotations for Pictures645

21.4.1 Annotations from Nearby Words646

21.4.2 Annotations from the Whole Image646

21.4.3 Predicting Correlated Words with Classifiers648

21.4.4 Names and Faces649

21.4.5 Generating Tags with Segments651

21.5 The State of the Art of Word Prediction654

21.5.1 Resources655

21.5.2 Comparing Methods655

21.5.3 Open Problems656

21.6 Notes659

Ⅶ BACKGROUND MATERIAL661

22 Optimization Techniques663

22.1 Linear Least-Squares Methods663

22.1.1 Normal Equations and the Pseudoinverse664

22.1.2 Homogeneous Systems and Eigenvalue Problems665

22.1.3 Generalized Eigenvalues Problems666

22.1.4 An Example：Fitting a Line to Points in a Plane666

22.1.5 Singular Value Decomposition667

22.2 Nonlinear Least-Squares Methods669

22.2.1 Newton’s Method：Square Systems of Nonlinear Equations670

22.2.2 Newton’s Method for Overconstrained Systems670

22.2.3 The Gauss-Newton and Levenberg-Marquardt Algorithms671

22.3 Sparse Coding and Dictionary Learning672

22.3.1 Sparse Coding672

22.3.2 Dictionary Learning673

22.3.3 Supervised Dictionary Learning675

22.4 Min-Cut/Max-Flow Problems and Combinatorial Optimization675

22.4.1 Min-Cut Problems676

22.4.2 Quadratic Pseudo-Boolean Functions677

22.4.3 Generalization to Integer Variables679

22.5 Notes682

Bibliography684

Index737

List of Algorithms760