His long experience in the field of puter and machine vision surpasses even the “big bang” in computer visionaround 25 years ago in the mid-80s when the Alvey Vision Conference UK andCVP
Trang 2Machine Vision: Theory, Algorithms,
Practicalities
Trang 3To my late father, Arthur Granville Davies, who passed on to me his appreciation of the beauties of mathematics and science.
To my wife, Joan, for love, patience, support, and inspiration.
To my children, Elizabeth, Sarah, and Marion, the music in my life.
To my grandson, Jasper, for reminding me of the carefree
joys of youth.
Trang 4Machine Vision: Theory, Algorithms,
Practicalities
Fourth Edition
E R DAVIES Department of Physics
Royal Holloway, University of London, Egham, Surrey, UK
AMSTERDAM • BOSTON • HEIDELBERG • LONDON
NEW YORK • OXFORD • PARIS • SAN DIEGO
SAN FRANCISCO • SINGAPORE • SYDNEY • TOKYO
Academic Press is an imprint of Elsevier
Trang 5Second edition 1997
Third edition 2005
Fourth edition 2012
Copyright r 2012 Elsevier Inc All rights reserved
No part of this publication may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher Details on how to seek permission, further information about the Publisher’s permissions policies and arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website:
Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.
Library of Congress Cataloging-in-Publication Data
A catalog record for this book is available from the Library of Congress
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
ISBN: 978-0-12-386908-1
For information on all Elsevier publications
visit our website at elsevierdirect.com
Typeset by MPS Limited, a Macmillan Company, Chennai, India
www.macmillansolutions.com
Printed and bound in United States of America
12 11 10 9 8 7 6 5 4 3 2 1
Trang 6Foreword xxi
Preface xxiii
About the Author xxvii
Acknowledgements xxix
Glossary of Acronyms and Abbreviations xxxiii
CHAPTER 1 Vision, the Challenge 1
1.1 Introduction—Man and His Senses 1
1.2 The Nature of Vision 2
1.2.1 The Process of Recognition 2
1.2.2 Tackling the Recognition Problem 4
1.2.3 Object Location 6
1.2.4 Scene Analysis 8
1.2.5 Vision as Inverse Graphics 9
1.3 From Automated Visual Inspection to Surveillance 10
1.4 What This Book is About 12
1.5 The Following Chapters 13
1.6 Bibliographical Notes 14
PART 1 LOW-LEVEL VISION 15 CHAPTER 2 Images and Imaging Operations 17
2.1 Introduction 18
2.1.1 Gray Scale Versus Color 19
2.2 Image Processing Operations 23
2.2.1 Some Basic Operations on Grayscale Images 24
2.2.2 Basic Operations on Binary Images 28
2.3 Convolutions and Point Spread Functions 32
2.4 Sequential Versus Parallel Operations 34
2.5 Concluding Remarks 36
2.6 Bibliographical and Historical Notes 36
2.7 Problems 36
CHAPTER 3 Basic Image Filtering Operations 38
3.1 Introduction 38
3.2 Noise Suppression by Gaussian Smoothing 40
3.3 Median Filters 43
3.4 Mode Filters 45
3.5 Rank Order Filters 52
v
Trang 73.6 Reducing Computational Load 54
3.7 Sharp Unsharp Masking 55
3.8 Shifts Introduced by Median Filters 56
3.8.1 Continuum Model of Median Shifts 57
3.8.2 Generalization to Grayscale Images 59
3.8.3 Problems with Statistics 60
3.9 Discrete Model of Median Shifts 62
3.10 Shifts Introduced by Mode Filters 65
3.11 Shifts Introduced by Mean and Gaussian Filters 67
3.12 Shifts Introduced by Rank Order Filters 68
3.12.1 Shifts in Rectangular Neighborhoods 69
3.13 The Role of Filters in Industrial Applications of Vision 74
3.14 Color in Image Filtering 74
3.15 Concluding Remarks 76
3.16 Bibliographical and Historical Notes 77
3.16.1 More Recent Developments 78
3.17 Problems 79
CHAPTER 4 Thresholding Techniques 82
4.1 Introduction 83
4.2 Region-Growing Methods 83
4.3 Thresholding 84
4.3.1 Finding a Suitable Threshold 85
4.3.2 Tackling the Problem of Bias in Threshold Selection 86
4.3.3 Summary 88
4.4 Adaptive Thresholding 88
4.4.1 The Chow and Kaneko Approach 91
4.4.2 Local Thresholding Methods 92
4.5 More Thoroughgoing Approaches to Threshold Selection 93
4.5.1 Variance-Based Thresholding 95
4.5.2 Entropy-Based Thresholding 96
4.5.3 Maximum Likelihood Thresholding 97
4.6 The Global Valley Approach to Thresholding 98
4.7 Practical Results Obtained Using the Global Valley Method 101
4.8 Histogram Concavity Analysis 106
4.9 Concluding Remarks 107
4.10 Bibliographical and Historical Notes 108
4.10.1 More Recent Developments 109
4.11 Problems 110
CHAPTER 5 Edge Detection 111
5.1 Introduction 112
5.2 Basic Theory of Edge Detection 113
Trang 85.3 The Template Matching Approach 115
5.4 Theory of 33 3 Template Operators 116
5.5 The Design of Differential Gradient Operators 117
5.6 The Concept of a Circular Operator 118
5.7 Detailed Implementation of Circular Operators 120
5.8 The Systematic Design of Differential Edge Operators 122
5.9 Problems with the Above Approach—Some Alternative Schemes 123
5.10 Hysteresis Thresholding 126
5.11 The Canny Operator 128
5.12 The Laplacian Operator 131
5.13 Active Contours 134
5.14 Practical Results Obtained Using Active Contours 137
5.15 The Level Set Approach to Object Segmentation 140
5.16 The Graph Cut Approach to Object Segmentation 141
5.17 Concluding Remarks 145
5.18 Bibliographical and Historical Notes 146
5.18.1 More Recent Developments 147
5.19 Problems 148
CHAPTER 6 Corner and Interest Point Detection 149
6.1 Introduction 150
6.2 Template Matching 150
6.3 Second-Order Derivative Schemes 151
6.4 A Median Filter-Based Corner Detector 153
6.4.1 Analyzing the Operation of the Median Detector 154
6.4.2 Practical Results 156
6.5 The Harris Interest Point Operator 158
6.5.1 Corner Signals and Shifts for Various Geometric Configurations 161
6.5.2 Performance with Crossing Points and Junctions 162
6.5.3 Different Forms of the Harris Operator 165
6.6 Corner Orientation 166
6.7 Local Invariant Feature Detectors and Descriptors 168
6.7.1 Harris Scale and Affine-Invariant Detectors and Descriptors 171
6.7.2 Hessian Scale and Affine-Invariant Detectors and Descriptors 173
6.7.3 The SIFT Operator 173
6.7.4 The SURF Operator 174
6.7.5 Maximally Stable Extremal Regions 176
6.7.6 Comparison of the Various Invariant Feature Detectors 177
Trang 96.8 Concluding Remarks 180
6.9 Bibliographical and Historical Notes 181
6.9.1 More Recent Developments 184
6.10 Problems 184
CHAPTER 7 Mathematical Morphology 185
7.1 Introduction 185
7.2 Dilation and Erosion in Binary Images 186
7.2.1 Dilation and Erosion 186
7.2.2 Cancellation Effects 186
7.2.3 Modified Dilation and Erosion Operators 187
7.3 Mathematical Morphology 187
7.3.1 Generalized Morphological Dilation 187
7.3.2 Generalized Morphological Erosion 188
7.3.3 Duality Between Dilation and Erosion 189
7.3.4 Properties of Dilation and Erosion Operators 190
7.3.5 Closing and Opening 193
7.3.6 Summary of Basic Morphological Operations 195
7.4 Grayscale Processing 197
7.4.1 Morphological Edge Enhancement 198
7.4.2 Further Remarks on the Generalization to Grayscale Processing 199
7.5 Effect of Noise on Morphological Grouping Operations 201
7.5.1 Detailed Analysis 203
7.5.2 Discussion 205
7.6 Concluding Remarks 205
7.7 Bibliographical and Historical Notes 206
7.7.1 More Recent Developments 207
7.8 Problem 208
CHAPTER 8 Texture 209
8.1 Introduction 209
8.2 Some Basic Approaches to Texture Analysis 213
8.3 Graylevel Co-occurrence Matrices 213
8.4 Laws’ Texture Energy Approach 217
8.5 Ade’s Eigenfilter Approach 220
8.6 Appraisal of the Laws and Ade Approaches 221
8.7 Concluding Remarks 223
8.8 Bibliographical and Historical Notes 223
8.8.1 More Recent Developments 224
Trang 10PART 2 INTERMEDIATE-LEVEL VISION 227
CHAPTER 9 Binary Shape Analysis 229
9.1 Introduction 230
9.2 Connectedness in Binary Images 230
9.3 Object Labeling and Counting 231
9.3.1 Solving the Labeling Problem in a More Complex Case 235
9.4 Size Filtering 238
9.5 Distance Functions and Their Uses 240
9.5.1 Local Maxima and Data Compression 243
9.6 Skeletons and Thinning 244
9.6.1 Crossing Number 247
9.6.2 Parallel and Sequential Implementations of Thinning 248 9.6.3 Guided Thinning 251
9.6.4 A Comment on the Nature of the Skeleton 251
9.6.5 Skeleton Node Analysis 251
9.6.6 Application of Skeletons for Shape Recognition 253
9.7 Other Measures for Shape Recognition 254
9.8 Boundary Tracking Procedures 257
9.9 Concluding Remarks 257
9.10 Bibliographical and Historical Notes 259
9.10.1 More Recent Developments 260
9.11 Problems 261
CHAPTER 10 Boundary Pattern Analysis 266
10.1 Introduction 266
10.2 Boundary Tracking Procedures 269
10.3 Centroidal Profiles 269
10.4 Problems with the Centroidal Profile Approach 270
10.4.1 Some Solutions 271
10.5 The (s, ψ) Plot 274
10.6 Tackling the Problems of Occlusion 276
10.7 Accuracy of Boundary Length Measures 279
10.8 Concluding Remarks 280
10.9 Bibliographical and Historical Notes 281
10.9.1 More Recent Developments 282
10.10 Problems 282
CHAPTER 11 Line Detection 284
11.1 Introduction 284
11.2 Application of the Hough Transform to Line Detection 285
11.3 The Foot-of-Normal Method 288
11.3.1 Application of the Foot-of-Normal Method 290
Trang 1111.4 Longitudinal Line Localization 290
11.5 Final Line Fitting 292
11.6 Using RANSAC for Straight Line Detection 293
11.7 Location of Laparoscopic Tools 297
11.8 Concluding Remarks 299
11.9 Bibliographical and Historical Notes 300
11.9.1 More Recent Developments 301
11.10 Problems 301
CHAPTER 12 Circle and Ellipse Detection 303
12.1 Introduction 304
12.2 Hough-Based Schemes for Circular Object Detection 305
12.3 The Problem of Unknown Circle Radius 308
12.3.1 Some Practical Results 310
12.4 The Problem of Accurate Center Location 311
12.4.1 A Solution Requiring Minimal Computation 313
12.5 Overcoming the Speed Problem 314
12.5.1 More Detailed Estimates of Speed 314
12.5.2 Robustness 315
12.5.3 Practical Results 316
12.5.4 Summary 317
12.6 Ellipse Detection 320
12.6.1 The Diameter Bisection Method 320
12.6.2 The Chord Tangent Method 322
12.6.3 Finding the Remaining Ellipse Parameters 323
12.7 Human Iris Location 325
12.8 Hole Detection 327
12.9 Concluding Remarks 327
12.10 Bibliographical and Historical Notes 328
12.10.1 More Recent Developments 330
12.11 Problems 331
CHAPTER 13 The Hough Transform and Its Nature 333
13.1 Introduction 333
13.2 The Generalized Hough Transform 334
13.3 Setting Up the Generalized Hough Transform—Some Relevant Questions 336
13.4 Spatial Matched Filtering in Images 336
13.5 From Spatial Matched Filters to Generalized Hough Transforms 337
13.6 Gradient Weighting Versus Uniform Weighting 339
13.6.1 Calculation of Sensitivity and Computational Load 339
13.7 Summary 342
13.8 Use of the GHT for Ellipse Detection 343
13.8.1 Practical Details 347
Trang 1213.9 Comparing the Various Methods 349
13.10 Fast Implementations of the Hough Transform 350
13.11 The Approach of Gerig and Klein 352
13.12 Concluding Remarks 353
13.13 Bibliographical and Historical Notes 354
13.13.1 More Recent Developments 356
13.14 Problems 357
CHAPTER 14 Pattern Matching Techniques 358
14.1 Introduction 359
14.2 A Graph-Theoretic Approach to Object Location 359
14.2.1 A Practical Example—Locating Cream Biscuits 363
14.3 Possibilities for Saving Computation 366
14.4 Using the Generalized Hough Transform for Feature Collation 369
14.4.1 Computational Load 370
14.5 Generalizing the Maximal Clique and Other Approaches 371
14.6 Relational Descriptors 373
14.7 Search 376
14.8 Concluding Remarks 377
14.9 Bibliographical and Historical Notes 378
14.9.1 More Recent Developments 380
14.10 Problems 381
PART 3 3-D VISION AND MOTION 387 CHAPTER 15 The Three-Dimensional World 389
15.1 Introduction 389
15.2 3-D Vision—the Variety of Methods 390
15.3 Projection Schemes for Three-Dimensional Vision 392
15.3.1 Binocular Images 393
15.3.2 The Correspondence Problem 396
15.4 Shape from Shading 398
15.5 Photometric Stereo 402
15.6 The Assumption of Surface Smoothness 405
15.7 Shape from Texture 407
15.8 Use of Structured Lighting 408
15.9 Three-Dimensional Object Recognition Schemes 410
15.10 Horaud’s Junction Orientation Technique 411
15.11 An Important Paradigm—Location of Industrial Parts 415
15.12 Concluding Remarks 417
15.13 Bibliographical and Historical Notes 419
15.13.1 More Recent Developments 420
15.14 Problems 421
Trang 13CHAPTER 16 Tackling the Perspective n-point Problem 424
16.1 Introduction 424
16.2 The Phenomenon of Perspective Inversion 425
16.3 Ambiguity of Pose under Weak Perspective Projection 427
16.4 Obtaining Unique Solutions to the Pose Problem 430
16.4.1 Solution of the Three-Point Problem 433
16.4.2 Using Symmetric Trapezia for Estimating Pose 434
16.5 Concluding Remarks 434
16.6 Bibliographical and Historical Notes 436
16.6.1 More Recent Developments 437
16.7 Problems 438
CHAPTER 17 Invariants and Perspective 439
17.1 Introduction 440
17.2 Cross-ratios: the “Ratio of Ratios” Concept 441
17.3 Invariants for Noncollinear Points 445
17.3.1 Further Remarks About the Five-Point Configuration 447
17.4 Invariants for Points on Conics 449
17.5 Differential and Semi-differential Invariants 452
17.6 Symmetric Cross-ratio Functions 454
17.7 Vanishing Point Detection 456
17.8 More on Vanishing Points 458
17.9 Apparent Centers of Circles and Ellipses 460
17.10 The Route to Face Recognition 462
17.10.1 The Face as Part of a 3-D Object 464
17.11 Perspective Effects in Art and Photography 466
17.12 Concluding Remarks 472
17.13 Bibliographical and Historical Notes 474
17.13.1 More Recent Developments 475
17.14 Problems 475
CHAPTER 18 Image Transformations and Camera Calibration 478
18.1 Introduction 479
18.2 Image Transformations 479
18.3 Camera Calibration 483
18.4 Intrinsic and Extrinsic Parameters 486
18.5 Correcting for Radial Distortions 488
18.6 Multiple View Vision 490
18.7 Generalized Epipolar Geometry 491
18.8 The Essential Matrix 492
18.9 The Fundamental Matrix 495
18.10 Properties of the Essential and Fundamental Matrices 496
18.11 Estimating the Fundamental Matrix 497
Trang 1418.12 An Update on the Eight-Point Algorithm 497
18.13 Image Rectification 498
18.14 3-D Reconstruction 499
18.15 Concluding Remarks 501
18.16 Bibliographical and Historical Notes 502
18.16.1 More Recent Developments 503
18.17 Problems 504
CHAPTER 19 Motion 505
19.1 Introduction 505
19.2 Optical Flow 506
19.3 Interpretation of Optical Flow Fields 509
19.4 Using Focus of Expansion to Avoid Collision 511
19.5 Time-to-Adjacency Analysis 513
19.6 Basic Difficulties with the Optical Flow Model 514
19.7 Stereo from Motion 515
19.8 The Kalman Filter 517
19.9 Wide Baseline Matching 519
19.10 Concluding Remarks 521
19.11 Bibliographical and Historical Notes 522
19.12 Problem 522
PART 4 TOWARD REAL-TIME PATTERN RECOGNITION SYSTEMS 523 CHAPTER 20 Automated Visual Inspection 525
20.1 Introduction 525
20.2 The Process of Inspection 527
20.3 The Types of Object to be Inspected 527
20.3.1 Food Products 528
20.3.2 Precision Components 528
20.3.3 Differing Requirements for Size Measurement 529
20.3.4 Three-Dimensional Objects 530
20.3.5 Other Products and Materials for Inspection 530
20.4 Summary: The Main Categories of Inspection 530
20.5 Shape Deviations Relative to a Standard Template 532
20.6 Inspection of Circular Products 533
20.7 Inspection of Printed Circuits 537
20.8 Steel Strip and Wood Inspection 538
20.9 Inspection of Products with High Levels of Variability 539
20.10 X-Ray Inspection 542
20.10.1 The Dual-Energy Approach to X-Ray Inspection 546
20.11 The Importance of Color in Inspection 546
Trang 1520.12 Bringing Inspection to the Factory 548
20.13 Concluding Remarks 549
20.14 Bibliographical and Historical Notes 550
20.14.1 More Recent Developments 552
CHAPTER 21 Inspection of Cereal Grains 553
21.1 Introduction 553
21.2 Case Study: Location of Dark Contaminants in Cereals 554
21.2.1 Application of Morphological and Nonlinear Filters to Locate Rodent Droppings 555
21.2.2 Problems with Closing 558
21.2.3 Ergot Detection Using the Global Valley Method 558
21.3 Case Study: Location of Insects 560
21.3.1 The Vectorial Strategy for Linear Feature Detection 560
21.3.2 Designing Linear Feature Detection Masks for Larger Windows 563
21.3.3 Application to Cereal Inspection 564
21.3.4 Experimental Results 564
21.4 Case Study: High-Speed Grain Location 566
21.4.1 Extending an Earlier Sampling Approach 566
21.4.2 Application to Grain Inspection 567
21.4.3 Summary 571
21.5 Optimizing the Output for Sets of Directional Template Masks 572
21.5.1 Application of the Formulae 573
21.5.2 Discussion 574
21.6 Concluding Remarks 575
21.7 Bibliographical and Historical Notes 575
21.7.1 More Recent Developments 576
CHAPTER 22 Surveillance 578
22.1 Introduction 579
22.2 Surveillance—The Basic Geometry 580
22.3 Foreground—Background Separation 584
22.3.1 Background Modeling 585
22.3.2 Practical Examples of Background Modeling 591
22.3.3 Direct Detection of the Foreground 593
22.4 Particle Filters 594
22.5 Use of Color Histograms for Tracking 600
22.6 Implementation of Particle Filters 604
22.7 Chamfer Matching, Tracking, and Occlusion 607
22.8 Combining Views from Multiple Cameras 609
Trang 1622.8.1 The Case of Nonoverlapping Fields of View 613
22.9 Applications to the Monitoring of Traffic Flow 614
22.9.1 The System of Bascle et al 614
22.9.2 The System of Koller et al .616
22.10 License Plate Location 619
22.11 Occlusion Classification for Tracking 621
22.12 Distinguishing Pedestrians by Their Gait 623
22.13 Human Gait Analysis 627
22.14 Model-Based Tracking of Animals 629
22.15 Concluding Remarks 631
22.16 Bibliographical and Historical Notes 632
22.16.1 More Recent Developments 634
22.17 Problem 635
CHAPTER 23 In-Vehicle Vision Systems 636
23.1 Introduction 637
23.2 Locating the Roadway 638
23.3 Location of Road Markings 640
23.4 Location of Road Signs 641
23.5 Location of Vehicles 645
23.6 Information Obtained by Viewing License Plates and Other Structural Features 647
23.7 Locating Pedestrians 651
23.8 Guidance and Egomotion 653
23.8.1 A Simple Path Planning Algorithm 656
23.9 Vehicle Guidance in Agriculture 656
23.9.1 3-D Aspects of the Task 660
23.9.2 Real-Time Implementation 661
23.10 Concluding Remarks 662
23.11 More Detailed Developments and Bibliographies Relating to Advanced Driver Assistance Systems 663
23.11.1 Developments in Vehicle Detection 664
23.11.2 Developments in Pedestrian Detection 666
23.11.3 Developments in Road and Lane Detection 668
23.11.4 Developments in Road Sign Detection 669
23.11.5 Developments in Path Planning, Navigation, and Egomotion 671
23.12 Problem 671
CHAPTER 24 Statistical Pattern Recognition 672
24.1 Introduction 673
24.2 The Nearest Neighbor Algorithm 674
24.3 Bayes’ Decision Theory 676
24.3.1 The Naive Bayes’ Classifier 678
Trang 1724.4 Relation of the Nearest Neighbor and Bayes’
Approaches 679
24.4.1 Mathematical Statement of the Problem 679
24.4.2 The Importance of the Nearest Neighbor Classifier 681
24.5 The Optimum Number of Features 681
24.6 Cost Functions and Error Reject Tradeoff 682
24.7 The Receiver Operating Characteristic 684
24.7.1 On the Variety of Performance Measures Relating to Error Rates 686
24.8 Multiple Classifiers 688
24.9 Cluster Analysis 691
24.9.1 Supervised and Unsupervised Learning 691
24.9.2 Clustering Procedures 692
24.10 Principal Components Analysis 695
24.11 The Relevance of Probability in Image Analysis 699
24.12 Another Look at Statistical Pattern Recognition: The Support Vector Machine 700
24.13 Artificial Neural Networks 701
24.14 The Back-Propagation Algorithm 705
24.15 MLP Architectures 708
24.16 Overfitting to the Training Data 709
24.17 Concluding Remarks 712
24.18 Bibliographical and Historical Notes 713
24.18.1 More Recent Developments 715
24.19 Problems 717
CHAPTER 25 Image Acquisition 718
25.1 Introduction 718
25.2 Illumination Schemes 719
25.2.1 Eliminating Shadows 721
25.2.2 Principles for Producing Regions of Uniform Illumination 724
25.2.3 Case of Two Infinite Parallel Strip Lights 726
25.2.4 Overview of the Uniform Illumination Scenario 729
25.2.5 Use of Line-Scan Cameras 730
25.2.6 Light Emitting Diode (LED) Sources 731
25.3 Cameras and Digitization 732
25.3.1 Digitization 734
25.4 The Sampling Theorem 735
25.5 Hyperspectral Imaging 738
25.6 Concluding Remarks 739
25.7 Bibliographical and Historical Notes 740
25.7.1 More Recent Developments 741
Trang 18CHAPTER 26 Real-Time Hardware and Systems Design
Considerations 742
26.1 Introduction 743
26.2 Parallel Processing 744
26.3 SIMD Systems 745
26.4 The Gain in Speed Attainable withN Processors 747
26.5 Flynn’s Classification 748
26.6 Optimal Implementation of Image Analysis Algorithms 750
26.6.1 Hardware Specification and Design 751
26.6.2 Basic Ideas on Optimal Hardware Implementation 752
26.7 Some Useful Real-Time Hardware Options 754
26.8 Systems Design Considerations 755
26.9 Design of Inspection Systems—the Status Quo 757
26.10 System Optimization 760
26.11 Concluding Remarks 761
26.12 Bibliographical and Historical Notes 763
26.12.1 General Background 763
26.12.2 Developments Since 2000 764
26.12.3 More Recent Developments 765
CHAPTER 27 Epilogue— Perspectives in Vision 767
27.1 Introduction 767
27.2 Parameters of Importance in Machine Vision 768
27.3 Tradeoffs 770
27.3.1 Some Important Tradeoffs 770
27.3.2 Tradeoffs for Two-Stage Template Matching 771
27.4 Moore’s Law in Action 772
27.5 Hardware, Algorithms, and Processes 773
27.6 The Importance of Choice of Representation 774
27.7 Past, Present, and Future 775
27.8 Bibliographical and Historical Notes 777
Appendix A Robust Statistics 778
References 796
Author Index 845
Subject Index 861
Trang 19Topics Covered in Application Case Studies
Trang 20Influences Impinging upon Integrated Vision
System Design
Trang 22Although computer vision is such a relatively young field of study, it has matured
immensely over the last 25 years or so—from well-constrained, targeted
applica-tions to systems that learn automatically from examples
Such progress over these 25 years has been spurred not least by mind-boggling
advances in vision and computational hardware, making possible simple tasks
that could take minutes on small images, now integrated as part of real-time
sys-tems that do far more in a fraction of a second on much larger images in a video
stream
This all means that the focus of research has been in a perpetual state of
change, marked by near-exponential advances and achievements, and witnessed
by the quality, and often quantity, of outstanding contributions to the field
pub-lished in key conferences and journals such as ICCV and PAMI These advances
are most clearly reflected by the growing importance of the application areas in
which the novel and real-time developments in computer vision have been applied
to or developed for Twenty-five years ago, industrial quality inspection and
sim-ple military applications ruled the waves, but the emphasis has since spread its
wings, some slowly and some like wildfire, to many more areas, for example,
from medical imaging and analysis to surveillance and, inevitably, complex
mili-tary and space applications
So how does Roy’s book reflect this shift? Naturally, there are many
funda-mental techniques that remain the same, and this book is a wonderful treasure
chest of tools that provides the fundamentals for any researcher and teacher
More modern and state-of-the-art methodologies are also covered in the book,
most of them pertinent to the topical application areas currently driving not only
the research agenda, but also the market forces In short, the book is a direct
reflection of the progress and key methodologies developed in computer vision
over the last 25 years and more
Indeed, while the third edition of this book was already an excellent,
success-ful, and internationally popular work, this fourth edition is greatly enhanced and
updated All its chapters have been substantially revised and brought up to date
by the inclusion of many new references covering advances in the subject made
even in the past year There are now also two entirely new chapters (to reflect the
great strides that have been made in the area of video analytics) on surveillance
and in-vehicle vision systems The latter is highly relevant to the coming era of
advanced driver assistance systems, and the former’s importance and role requires
no emphasis in this day and age where so many resources are dedicated to
crimi-nal and terrorist activity monitoring and prevention
The material in the book is written in a way that is both approachable and
didactic It is littered with examples and algorithms I am sure that this volume
will be welcomed by a great many students and workers in computer and machine
vision, including practitioners in academia and industry—from beginners who are
xxi
Trang 23starting out in the subject to advanced researchers and workers who need to gaininsight into video analytics I will also welcome it personally, for use by my ownundergraduate and postgraduate students, and will value its presence on my book-shelf as an up-to-date reference on this important subject.
Finally, I am very happy to go on record as saying that Roy is the right person
to have produced this substantial work His long experience in the field of puter and machine vision surpasses even the “big bang” in computer visionaround 25 years ago in the mid-80s when the Alvey Vision Conference (UK) andCVPR (USA) were only inchoates of what they have become today and reachesback to when ICPR and IAPR began to be dominated by image processing in thelate 70s
com-September 2011Majid MirmehdiUniversity of Bristol, UK
Trang 24PREFACE TO THE FOURTH EDITION
The first edition came out in 1990, and was welcomed by many researchers and
practitioners However, in the subsequent two decades, the subject moved on at a
rapidly accelerating rate, and many topics that hardly deserved a mention in the
first edition had to be solidly incorporated in subsequent editions It seemed
par-ticularly important to bring in significant amounts of new material on
mathemati-cal morphology, 3-D vision, invariance, motion analysis, object tracking, artificial
neural networks, texture analysis, X-ray inspection, foreign object detection, and
robust statistics There are thus new chapters or appendices on these topics, and
they have been carefully integrated with the existing material The greater
propor-tion of the new material has been included in Parts 3 and 4 So great has been the
growth in work on 3-D vision and its applications that the original single chapter
on 3-D vision had to be expanded into the set offive chapters on 3-D vision and
motion forming Part 3, together with a further two chapters on surveillance and
in-vehicle vision systems in Part 4 Indeed, these changes have been so radical
that the title of the book has had to be modified to reflect them At this stage,
Part 4 encompasses such a range of chapters—covering applications and the
com-ponents needed for constructing real-time visual pattern recognition systems—
that it is difficult to produce a logical ordering for them: notably, the topics
interact with each other at a variety of different levels—theory, algorithms,
meth-odologies, practicalities, design constraints, and so on However, this should not
matter in practice, as the reader will be exposed to the essential richness of the
subject, and his/her studies should be amply rewarded by increased understanding
and capability
It is worth remarking that, at this point in time, computer vision has attained a
level of maturity that has made it substantially more rigorous, reliable, generic,
and—in the light of the improved hardware facilities now available for its
imple-mentation (not least, FPGA and GPU types of solution)—capable of real-time
performance This means that workers are more than ever before using it in
seri-ous applications, and with fewer practical difficulties It is intended that this
edi-tion of the book will reflect this radically new and exciting state of affairs at a
fundamental level
A typical final-year undergraduate course on vision for electronic engineering
or computer science students might include much of the work of Chapters 1 10
and 14, 15, plus a selection of sections from other chapters, according to
require-ments For MSc or PhD research students, a suitable lecture course might go on
xxiii
Trang 25to cover Part 3 in depth, including several of the chapters in Part 4,1with manypractical exercises being undertaken on an image analysis system Here, muchwill depend on the research program being undertaken by each individual student.
At this stage, the text will have to be used more as a handbook for research, andindeed, one of the prime aims of the volume is to act as a handbook for theresearcher and practitioner in this important area
As mentioned in the original Preface, this book leans heavily on experience Ihave gained from working with postgraduate students: in particular, I would like toexpress my gratitude to Mark Edmonds, Simon Barker, Daniel Celano, DarrelGreenhill, Derek Charles, Mark Sugrue, and Georgios Mastorakis, all of whom have
in their own ways helped to shape my view of the subject In addition, it is a specialpleasure to recall very many rewarding discussions with my colleagues Barry Cook,Zahid Hussain, Ian Hannah, Dev Patel, David Mason, Mark Bateman, Tieying Lu,Adrian Johnstone, and Piers Plummer, the last two named having been particularlyprolific in generating hardware systems for implementing my research group’svision algorithms Next, I am immensely grateful to Majid Mirmehdi for readingmuch of the manuscript and making insightful comments and valuable suggestions.Finally, I am indebted to Tim Pitts of Elsevier Science for his help and encourage-ment, without which this fourth edition might never have been completed
PREFACE TO THE FIRST EDITION (1990)
Over the past 30 years or so, machine vision has evolved into a mature subjectembracing many topics and applications: these range from automatic (robot)assembly to automatic vehicle guidance, from automatic interpretation of docu-ments to verification of signatures, and from analysis of remotely sensed images
to checking of fingerprints and human blood cells; currently, automated visualinspection is undergoing very substantial growth, necessary improvements in
1 The importance of the appendix on robust statistics should not be underestimated once one gets onto serious work, although this will probably be outside the restrictive environment of an under- graduate syllabus.
Trang 26quality, safety and cost-effectiveness being the stimulating factors With so much
ongoing activity, it has become a difficult business for the professional to keep up
with the subject and with relevant methodologies: in particular, it is difficult to
distinguish accidental developments from genuine advances It is the purpose of
this book to provide background in this area
The book was shaped over a period of 10 12 years, through material I have
given on undergraduate and postgraduate courses at London University, and
con-tributions to various industrial courses and seminars At the same time, my own
investigations coupled with experience gained while supervising PhD and
post-doctoral researchers helped to form the state of mind and knowledge that is now
set out here Certainly it is true to say that if I had had this book 8, 6, 4, or even
2 years ago, it would have been of inestimable value to myself for solving
practi-cal problems in machine vision It is therefore my hope that it will now be of use
to others in the same way Of course, it has tended to follow an emphasis that is
my own—and in particular one view of one path toward solving automated visual
inspection and other problems associated with the application of vision in
indus-try At the same time, although there is a specialism here, great care has been
taken to bring out general principles—including many applying throughout the
field of image analysis The reader will note the universality of topics such as
noise suppression, edge detection, principles of illumination, feature recognition,
Bayes’ theory, and (nowadays) Hough transforms However, the generalities lie
deeper than this The book has aimed to make some general observations and
messages about the limitations, constraints, and tradeoffs to which vision
algo-rithms are subject Thus, there are themes about the effects of noise, occlusion,
distortion and the need for built-in forms of robustness (as distinct from less
suc-cessful ad hoc varieties and those added on as an afterthought); there are also
themes about accuracy, systematic design, and the matching of algorithms and
architectures Finally, there are the problems of setting up lighting schemes
which must be addressed in complete systems, yet which receive scant attention
in most books on image processing and analysis These remarks will indicate that
the text is intended to be read at various levels—a factor that should make it of
more lasting value than might initially be supposed from a quick perusal of the
Contents
Of course, writing a text such as this presents a great difficulty in that it is
necessary to be highly selective: space simply does not allow everything in a
sub-ject of this nature and maturity to be dealt with adequately between two covers
One solution might be to dash rapidly through the whole area mentioning
every-thing that comes to mind, but leaving the reader unable to understand anyevery-thing in
detail or toachieve anything having read the book However, in a practical
sub-ject of this nature, this seemed to me a rather worthless extreme It is just possible
that the emphasis has now veered too much in the opposite direction, by coming
down to practicalities (detailed algorithms, details of lighting schemes, and so
on): individual readers will have to judge this for themselves On the other hand,
an author has to be true to himself and my view is that it is better for a reader or
Trang 27student to have mastered a coherent series of topics than to have a mish-mash ofinformation that he is later unable to recall with any accuracy This, then, is myjustification for presenting this particular material in this particular way and forreluctantly omitting from detailed discussion such important topics as textureanalysis, relaxation methods, motion, and optical flow.
As for the organization of the material, I have tried to make the early part ofthe book lead into the subject gently, giving enough detailed algorithms (espe-cially in Chapters 2 and 6) to provide a sound feel for the subject—includingespecially vital, and in their own way quite intricate, topics such as connectedness
in binary images Hence, Part 1 provides the lead-in, although it is not alwaystrivial material and indeed some of the latest research ideas have been brought in(e.g., on thresholding techniques and edge detection) Part 2 gives much of themeat of the book Indeed, the (book) literature of the subject currently has a sig-nificant gap in the area of intermediate-level vision; while high-level vision (AI)topics have long caught the researcher’s imagination, intermediate-level visionhas its own difficulties which are currently being solved with great success (notethat the Hough transform, originally developed in 1962, and by many thought to
be a very specialist topic of rather esoteric interest, is arguably only now cominginto its own) Part 2 and the early chapters of Part 3 aim to make this clear, whilePart 4 gives reasons why this particular transform has become so useful As awhole, Part 3 aims to demonstrate some of the practical applications of the basicwork covered earlier in the book, and to discuss some of the principles underlyingimplementation: it is here that chapters on lighting and hardware systems will befound As there is a limit to what can be covered in the space available, there is acorresponding emphasis on the theory underpinning practicalities Probably, this
is a vital feature, since there are many applications of vision both in industry andelsewhere, yet listing them and their intricacies risks dwelling on interminabledetail, which some might find insipid; furthermore, detail has a tendency to daterather rapidly Although the book could not cover 3-D vision in full (this topicwould easily consume a whole volume in its own right), a careful overview ofthis complex mathematical and highly important subject seemed vital It is there-fore no accident that Chapter 16 is the longest in the book Finally, Part 4 asksquestions about the limitations and constraints of vision algorithms and answersthem by drawing on information and experience from earlier chapters It is tempt-ing to call the last chapter the Conclusion However, in such a dynamic subjectarea, any such temptation has to be resisted, although it has still been possible todraw a good number of lessons on the nature and current state of the subject.Clearly, this chapter presents a personal view but I hope it is one that readers willfind interesting and useful
Trang 28Roy Davies is Emeritus Professor of Machine Vision at Royal Holloway,
University of London He has worked on many aspects of vision, from feature
detection and noise suppression to robust pattern matching and real-time
imple-mentations of practical vision tasks His interests include automated visual
inspec-tion, surveillance, vehicle guidance, and crime detection He has published more
than 200 papers and three books—Machine Vision: Theory, Algorithms,
Practicalities (1990), Electronics, Noise and Signal Recovery (1993), and Image
Processing for the Food Industry (2000); the first of these has been widely used
internationally for more than 20 years, and is now out in this much enhanced
fourth edition Roy is a Fellow of the IoP and the IET, and a Senior Member of
the IEEE He is on the Editorial Boards of Real-Time Image Processing, Pattern
Recognition Letters, Imaging Science, and IET Image Processing He holds a DSc
at the University of London: he was awarded BMVA Distinguished Fellow in
2005 and Fellow of the International Association of Pattern Recognition in 2008
xxvii
Trang 30The author would like to credit the following sources for permission to reproduce
tables, figures and extracts of text from earlier publications:
Elsevier
For permission to reprint portions of the following papers from Image and Vision
Computing as text in Chapters 5 and 14; as Tables 5.1 5.5; and as Figures 3.29,
5.2, 14.1, 14.2, 14.6:
Davies (1984b, 1987c)
Davies, E.R (1991) Image and Vision Computing9, 252 261
For permission to reprint portions of the following paper from Pattern
Recognition as text in Chapter 9; and as Figure 9.11:
Davies and Plummer (1981)
For permission to reprint portions of the following papers from Pattern
Recognition Letters as text in Chapters 3, 5, 11 14, 21, 24; as Tables 3.2; 12.3;
13.1; and as Figures 3.6, 3.8, 3.10; 5.1, 5.3; 11.1, 11.2a, 11.3b; 12.4, 12.5, 12.6,
12.7 12.10; 13.1, 13.3 13.11; 21.3, 21.6:
Davies (1986a,b; 1987a,e,f; 1988b,c,e,f; 1989a)
Davies et al (2003a)
For permission to reprint portions of the following paper from Signal Processing
as text in Chapter 3; and as Figures 3.15 3.20:
Davies (1989b)
For permission to reprint portions of the following paper from Advances in
Imaging and Electron Physics as text in Chapter 3:
Davies (2003c)
For permission to reprint portions of the following article from Encyclopedia of
Physical Science and Technology as Figures 9.9, 9.12, 10.1, 10.4:
Davies, E.R (1987) Visual inspection, automatic (robotics) In: Meyers, R.A
(ed.) Encyclopedia of Physical Science and Technology, Vol 14 Academic
Press, San Diego, pp 360 377
The Committee of the Alvey Vision Club
For permission to reprint portions of the following paper as text in Chapter 14;
and as Figures 14.1, 14.2, 14.6:
Davies, E.R (1988) An alternative to graph matching for locating objects
from their salient features Proc 4th Alvey Vision Conf., Manchester (31
August 2 September), pp 281 286
xxix
Trang 31CEP Consultants Ltd (Edinburgh)
For permission to reprint portions of the following paper as text in Chapter 20:Davies, E.R (1987) Methods for the rapid inspection of food products andsmall parts In: McGeough, J.A (ed.) Proc 2nd Int Conf on Computer-AidedProduction Engineering, Edinburgh (13 15 April), pp 105 110
For permission to reprint portions of the following paper as text in Chapter 3; and
Davies (1985; 1988a; 1997b; 1999f; 2000b,c; 2005; 2008b)
Davies, E.R (1997) Algorithms for inspection: constraints, tradeoffs and thedesign process IEE Digest no 1997/041, Colloquium on IndustrialInspection, IEE (10 Feb.), pp 6/1 5
Sugrue and Davies (2007)
Mastorakis and Davies (2011)
Davies et al (1998a)
Davies and Johnstone (1989)
IFS Publications Ltd
For permission to reprint portions of the following paper as text in Chapters 12,20; and as Figures 12.1, 12.2, 20.5:
Davies (1984c)
Trang 32The Council of the Institution of Mechanical Engineers
For permission to reprint portions of the following paper as text in Chapter 26;
and as Tables 26.1, 26.2:
Davies and Johnstone (1986)
MCB University Press (Emerald Group)
For permission to reprint portions of the following paper as Figure 20.6:
Patel et al (1995)
The Royal Photographic Society
For permission to reprint portions of the following papers1 as text in Chapter 3;
as Table 3.4; and as Figures 3.12, 3.13, 3.25 3.28:
Davies, E.R (2003) Design of object location algorithms and their use for
food and cereals inspection Chapter 15 in Graves, M and Batchelor, B.G
(eds.) Machine Vision Techniques for Inspecting Natural Products
Springer-Verlag, pp 393 420
Peter Stevens Photography
For permission to reprint a photograph as Figure 3.12(a)
F.H Sumner
For permission to reprint portions of the following article from State of the Art
Report: Supercomputer Systems Technology as text in Chapter 9; and as
Figure 9.4:
Davies, E.R (1982) Image processing In: Sumner, F.H (ed.) State of the Art
Report: Supercomputer Systems Technology Pergamon Infotech, Maidenhead,
pp 223 244
1 See also the Maney website: www.maney.co.uk/journals/ims
Trang 33World Scientific
For permission to reprint portions of the following book as text in Chapters 7, 21,
22, 23, 26; and as Figures 7.1 7.4, 21.4, 22.20, 23.15, 23.16, 26.3:
Davies (2000a)
Royal Holloway, University of London
For permission to reprint extracts from the following examination questions, inally written by E.R Davies:
orig-EL385/97/2; EL333/98/2; EL333/99/2, 3, 5, 6; EL333/01/2, 4 6;
Trang 341-D one dimension/one-dimensional
2-D two dimensions/two-dimensional
3-D three dimensions/three-dimensional
3DPO 3-D part orientation system
ACM Association for Computing Machinery (USA)
ADAS advanced driver assistance system
ADC analog to digital converter
AI artificial intelligence
ANN artificial neural network
APF auxiliary particle filter
ASCII American Standard Code for Information Interchange
ASIC application-specific integrated circuit
ATM automated teller machine
AUC area under curve
AVI audio video interleave
BCVM between-class variance method
BetaSAC beta [distribution] sampling consensus
BMVA British Machine Vision Association
BRAM block of RAM
BRDF bidirectional reflectance distribution function
CAD computed-aided design
CAM computer-aided manufacture
CCD charge-coupled device
CCTV closed-circuit television
CDF cumulative distribution function
CIM computer integrated manufacture
CLIP cellular logic image processor
CPU central processing unit
DCSM distinct class based splitting measure
DET Beaudet determinant operator
DEXA dual-emission X-ray absorptiometry
DG differential gradient
DN Dreschler Nagel corner detector
DoF degree of freedom
DoG difference of Gaussians
DSP digital signal processor
EM expectation maximization
EURASIP European Association for Signal Processing
FAST features from accelerated segment test
FFT fast Fourier transform
xxxiii
Trang 35FN false negative
fnr false negative rate
FoE focus of expansion
FoV field of view
FP false positive
FPGA field programmable gate array
FPP full perspective projection
fpr false positive rate
GHT generalized Hough transform
GLOH gradient location and orientation histogram
GMM Gaussian mixture model
GPS global positioning system
GPU graphics processing unit
GroupSAC group sampling consensus
GVM global valley method
HOG histogram of orientated gradients
HSI hue, saturation, intensity
HT Hough transform
IBR intensity extrema-based region detector
IDD integrated directional derivative
IEE Institution of Electrical Engineers (UK)
IEEE Institute of Electrical and Electronics Engineers (USA)IET Institution of Engineering and Technology (UK)ILW iterated likelihood weighting
IMechE Institution of Mechanical Engineers (UK)
IMPSAC importance sampling consensus
ISODATA iterative self-organizing data analysis
JPEG/JPG Joint Photographic Experts Group
k-NN k-nearest neighbor
KR Kitchen Rosenfeld corner detector
LED light emitting diode
LFF local-feature-focus method
LIDAR light detection and ranging
LMedS least median of squares
LoG Laplacian of Gaussian
LS least squares
LUT lookup table
MAP maximum a posteriori
MDL minimum description length
MIMD multiple instruction stream, multiple data streamMIPS millions of instructions per second
MISD multiple instruction stream, single data stream
MLP multi-layer perceptron
Trang 36MoG mixture of Gaussians
MP microprocessor
MSER maximally stable extremal region
NAPSAC n adjacent points sample consensus
NIR near infra-red
NN nearest neighbor
OCR optical character recognition
PC personal computer
PCA principal components analysis
PCB printed circuit board
PE processing element
PnP perspectiven-point
PR pattern recognition
PROSAC progressive sample consensus
PSF point spread function
RAM random access memory
RANSAC random sample consensus
RGB red, green, blue
RHT randomized Hough transform
RKHS reproducible kernel Hilbert space
RMS root mean square
ROC receiver operating characteristic
RoI region of interest
RPS Royal Photographic Society (UK)
SFOP scale-invariant feature operator
SIAM Society of Industrial and Applicative Mathematics
SIFT scale-invariant feature transform
SIMD single instruction stream, multiple data stream
SIR sampling importance resampling
SIS sequential importance sampling
SISD single instruction stream, single data stream
SOC sorting optimization curve
SOM self-organizing map
SPIE Society of Photo-optical Instrumentation Engineers
SPR statistical pattern recognition
STA spatiotemporal attention [neural network]
SURF speeded-up robust features
SUSAN smallest univalue segment assimilating nucleus
SVM support vector machine
Trang 37TP true positive
tpr true positive rate
TV television
ULUT universal lookup table
USEF unit step edge function
VLSI very large scale integrationVMF vector median filter
VP vanishing point
WPP weak perspective projection
ZH Zuniga Haralick corner detector
Trang 381 Vision, the Challenge
Of the five senses—vision, hearing, smell, taste, and touch—vision is undoubtedly
the one that man has come to depend upon above all others, and indeed the
one that provides most of the data he receives Not only do the input pathways
from the eyes provide megabits of information at each glance but the data rates for
continuous viewing probably exceed 10 megabits per second (mbit/s) However,
much of this information is redundant and is compressed by the various layers of
the visual cortex, so that the higher centers of the brain have to interpret abstractly
only a small fraction of the data Nonetheless, the amount of information the higher
centers receive from the eyes must be at least two orders of magnitude greater than
all the information they obtain from the other senses
Another feature of the human visual system is the ease with which interpretation
is carried out We see a scene as it is—trees in a landscape, books on a desk,
widgets in a factory No obvious deductions are needed and no overt effort is
required to interpret each scene: in addition, answers are effectively immediate
and are normally available within a tenth of a second Just now and again some
doubt arises—e.g a wire cube might be “seen” correctly or inside out This and
a host of other optical illusions are well known, although for the most part we
can regard them as curiosities—irrelevant freaks of nature Somewhat surprisingly,
illusions are quite important, since they reflect hidden assumptions that the brain is
making in its struggle with the huge amounts of complex visual data it is receiving
We have to pass by this story here (although it resurfaces now and again in various
parts of this book) However, the important point is that we are for the most part
unaware of the complexities of vision Seeing is not a simple process: it is just that
vision has evolved over millions of years, and there was no particular advantage
in evolution giving us any indication of the difficulties of the task (if anything,
Trang 39to have done so would have cluttered our minds with irrelevant information andslowed our reaction times).
In the present day and age, man is trying to get machines to do much of hiswork for him For simple mechanistic tasks this is not particularly difficult, butfor more complex tasks the machine must be given the sense of vision Effortshave been made to achieve this, sometimes in modest ways, for well over 30 years
At first, schemes were devised for reading, for interpreting chromosome images,and so on, but when such schemes were confronted with rigorous practical tests,the problems often turned out to be more difficult Generally, researchers react
to finding that apparent “trivia” are getting in the way by intensifying their effortsand applying great ingenuity, and this was certainly so with early efforts at visionalgorithm design Hence, it soon became evident that the task really is a complexone, in which numerous fundamental problems confront the researcher, and theease with which the eye can interpret scenes turned out to be highly deceptive
Of course, one of the ways in which the human visual system gains over themachine is that the brain possesses more than 1010 cells (or neurons), some ofwhich have well over 10,000 contacts (or synapses) with other neurons If eachneuron acts as a type of microprocessor, then we have an immense computer inwhich all the processing elements can operate concurrently Taking the largestsingle man-made computer to contain several hundred million rather modestprocessing elements, the majority of the visual and mental processing tasks thatthe eyebrain system can perform in a flash have no chance of being performed
by present-day man-made systems Added to these problems of scale, there isthe problem of how to organize such a large processing system, and also how toprogram it Clearly, the eyebrain system is partly hard-wired by evolution butthere is also an interesting capability to program it dynamically by training duringactive use This need for a large parallel processing system with the attendantcomplex control problems shows that machine vision must indeed be one of themost difficult intellectual problems to tackle
So what are the problems involved in vision that make it apparently so easyfor the eye, yet so difficult for the machine? In the next few sections an attempt
is made to answer this question
1.2.1 The Process of Recognition
This section illustrates the intrinsic difficulties of implementing machine vision,starting with an extremely simple example—that of character recognition Considerthe set of patterns shown inFig 1.1(a) Each pattern can be considered as a set of
25 bits of information, together with an associated class indicating its interpretation
In each case imagine a computer learning the patterns and their classes by rote.Then any new pattern may be classified (or “recognized”) by comparing it with
Trang 40this previously learnt “training set,” and assigning it to the class of the nearest
pattern in the training set Clearly, test pattern (1) (Fig 1.1(b)) will be allotted to
class U on this basis Chapter 24 shows that this method is a simple form of the
nearest-neighbor approach to pattern recognition
The scheme outlined above seems straightforward and is indeed highly effective,
even being able to cope with situations where distortions of the test patterns occur or
where noise is present: this is illustrated by test patterns (2) and (3) However, this
approach is not always foolproof First, there are situations where distortions or noise
are excessive, so errors of interpretation arise Second, there are situations where
Some simple 25-bit patterns and their recognition classes used to illustrate some of
the basic problems of recognition: (a) training set patterns (for which the known classes
are indicated); (b) test patterns