لایب سان

This chapter presents an overview of the problems associated with character recognition. It includes a brief description of the history of OCR (optical character recognition), the extensive efforts involved to make it work, the recent international activities that have stimulated the growth of research in handwriting recognition and document analysis, a short summary of the topics to be discussed in the other chapters, and a list of relevant references. 1.1 GENERATION AND RECOGNITION OF CHARACTERS Most people learn to read and write during their first few years of education. By the time they have grown out of childhood, they have already acquired very good reading and writing skills, including the ability to read most texts, whether they are printed in different fonts and styles, or handwritten neatly or sloppily. Most people have no problem in reading the following: light prints or heavy prints; upside down prints; advertisements in fancy font styles; characters with flowery ornaments and missing parts; and even characters with funny decorations, stray marks, broken, or fragmented parts; misspelled words; and artistic and figurative designs. At times, the characters and words may appear rather distorted and yet, by experience and by context, most people can still figure them out. On the contrary, despite more than five decades of intensive research, the reading skill of the computer is still way behind that of human

CONTENTS Preface xiii Acknowledgments xvii List of Figures xix List of Tables xxvii Acronyms xxix 1 Introduction: Character Recognition, Evolution, and Development 1 1.1 Generation and Recognition of Characters 1 1.2 History of OCR 2 1.3 Development of New Techniques 3 1.4 Recent Trends and Movements 3 1.5 Organization of the Remaining Chapters 3 References 4 2 Tools for Image Preprocessing 5 2.1 Generic Form-Processing System 5 2.2 A Stroke Model for Complex Background Elimination 8 vii viii CONTENTS 2.2.1 Global Gray Level Thresholding 9 2.2.2 Local Gray Level Thresholding 11 2.2.3 Local Feature Thresholding—Stroke-Based Model 12 2.2.4 Choosing the Most Efficient Character Extraction Method 15 2.2.5 Cleaning Up Form Items Using Stroke-Based Model 19 2.3 A Scale-Space Approach for Visual Data Extraction 21 2.3.1 Image Regularization 22 2.3.2 Data Extraction 24 2.3.3 Concluding Remarks 29 2.4 Data Preprocessing 30 2.4.1 Smoothing and Noise Removal 30 2.4.2 Skew Detection and Correction 32 2.4.3 Slant Correction 34 2.4.4 Character Normalization 36 2.4.5 Contour Tracing/Analysis 41 2.4.6 Thinning 45 2.5 Chapter Summary 50 References 51 3 Feature Extraction, Selection, and Creation 54 3.1 Feature Extraction 54 3.1.1 Moments 55 3.1.2 Histogram 58 3.1.3 Direction Features 59 3.1.4 Image Registration 64 3.1.5 Hough Transform 68 3.1.6 Line-Based Representation 70 3.1.7 Fourier Descriptors 73 3.1.8 Shape Approximation 76 3.1.9 Topological Features 78 3.1.10 Linear Transforms 79 3.1.11 Kernels 86 3.2 Feature Selection for Pattern Classification 90 3.2.1 Review of Feature Selection Methods 90 3.3 Feature Creation for Pattern Classification 104 3.3.1 Categories of Feature Creation 104 3.3.2 Review of Feature Creation Methods 105 3.3.3 Future Trends 118 CONTENTS ix 3.4 Chapter Summary 120 References 120 4 Pattern Classification Methods 129 4.1 Overview of Classification Methods 129 4.2 Statistical Methods 131 4.2.1 Bayes Decision Theory 131 4.2.2 Parametric Methods 132 4.2.3 Nonparametric Methods 138 4.3 Artificial Neural Networks 142 4.3.1 Single-Layer Neural Network 144 4.3.2 Multilayer Perceptron 148 4.3.3 Radial Basis Function Network 152 4.3.4 Polynomial Network 155 4.3.5 Unsupervised Learning 156 4.3.6 Learning Vector Quantization 160 4.4 Support Vector Machines 162 4.4.1 Maximal Margin Classifier 163 4.4.2 Soft Margin and Kernels 165 4.4.3 Implementation Issues 166 4.5 Structural Pattern Recognition 171 4.5.1 Attributed String Matching 172 4.5.2 Attributed Graph Matching 174 4.6 Combining Multiple Classifiers 179 4.6.1 Problem Formulation 180 4.6.2 Combining Discrete Outputs 181 4.6.3 Combining Continuous Outputs 183 4.6.4 Dynamic Classifier Selection 190 4.6.5 Ensemble Generation 190 4.7 A Concrete Example 194 4.8 Chapter Summary 197 References 197 5 Word and String Recognition 204 5.1 Introduction 204 5.2 Character Segmentation 206 5.2.1 Overview of Dissection Techniques 207 5.2.2 Segmentation of Handwritten Digits 210 x CONTENTS 5.3 Classification-Based String Recognition 214 5.3.1 String Classification Model 214 5.3.2 Classifier Design for String Recognition 220 5.3.3 Search Strategies 227 5.3.4 Strategies for Large Vocabulary 234 5.4 HMM-Based Recognition 237 5.4.1 Introduction to HMMs 237 5.4.2 Theory and Implementation 238 5.4.3 Application of HMMs to Text Recognition 243 5.4.4 Implementation Issues 244 5.4.5 Techniques for Improving HMMs’ Performance 247 5.4.6 Summary to HMM-Based Recognition 250 5.5 Holistic Methods for Handwritten Word Recognition 250 5.5.1 Introduction to Holistic Methods 251 5.5.2 Overview of Holistic Methods 255 5.5.3 Summary to Holistic Methods 256 5.6 Chapter Summary 256 References 257 6 Case Studies 263 6.1 Automatically Generating Pattern Recognizers with Evolutionary Computation 263 6.1.1 Motivation 264 6.1.2 Introduction 264 6.1.3 Hunters and Prey 266 6.1.4 Genetic Algorithm 271 6.1.5 Experiments 272 6.1.6 Analysis 280 6.1.7 Future Directions 281 6.2 Offline Handwritten Chinese Character Recognition 282 6.2.1 Related Works 283 6.2.2 System Overview 285 6.2.3 Character Normalization 286 6.2.4 Direction Feature Extraction 289 6.2.5 Classification Methods 293 6.2.6 Experiments 293 6.2.7 Concluding Remarks 301 CONTENTS xi 6.3 Segmentation and Recognition of Handwritten Dates on Canadian Bank Cheques 301 6.3.1 Introduction 302 6.3.2 System Architecture 303 6.3.3 Date Image Segmentation 303 6.3.4 Date Image Recognition 308 6.3.5 Experimental Results 315 6.3.6 Concluding Remarks 317 References 317 Index 321

برای ارسال نظر لطفا وارد شوید یا ثبت نام کنید