cheriet, kharma, liu, suen - character recognition systems. a guide for students and practitioners

[...]... made by classifiers by digit Accuracies (%) of 4-orientation and 8-direction chaincode feature by zoning and by blurring on ETL9B database Accuracies (%) of 4-orientation and 8-direction chaincode feature by zoning and by blurring on CASIA database Accuracies (%) of 8-direction NCFE (discrete and continuous, denoted by ncf-d and ncf-c) and gradient feature (binary and gray, denoted by grd-b and grd-g)... 16-direction features Accuracies (%) of 8-direction, 12-direction, and 16-direction features on ETL9B database Accuracies (%) of 8-direction, 12-direction, and 16-direction features on CASIA database Accuracies (%) of various normalization methods on ETL9B database Accuracies (%) of various normalization methods on CASIA database Accuracies (%) of MQDF with variable number k of principal eigenvectors and LVQ... search and parallel search Operator set for the GP [38] Test accuracies (%) of individual classifiers, oracle, and DCS Accuracies (%) of combination at abstract level: plurality vote and trained rules Accuracies (%) of combination at measurement level: fixed and trained rules Affinity-bits based control of the merger mechanism Maximum training and validation accuracies for the binary classifiers Percentage agreement... procedures, and testing of real-life databases are also provided In real-world applications, documents consist of words or character strings rather than elementary units of isolated characters However, word and character string recognition processes involve character segmentation, character recognition, and contextual analyses Chapter 5 gives an overview of word and string recognition methods, and techniques... example agent Agent A1 and its partition of feature space Agent A2 and its partition of feature space Agent A3 and its partition of feature space Agent A4 and its partition of feature space (a) Typical run of the ME trial; (b) typical run of the SGA trial (a) Best run of the LAC trial accuracy; (b) best run of the LAC trial complexity (a) Best run of the IAC trial accuracy; (b) best run of the IAC... in a digital text image This chapter also introduces the essential smoothing and noise removal techniques and key methods of normalizing the digitized characters, including the detection and correction of slants and skews, and the normalization of character sizes and positions As well, methods of extracting character contours and skeletons are presented with appropriate illustrations All characters and. .. four approaches to feature selection Feature subset evaluation criteria Search space Forward search The basic schemes of sequential search and parallel search A taxonomy of feature selection search strategies The space of feature creations methods Automatic operator (feature) creation procedure [110] A feature detection operator [105] A hit-miss matrix and its application to a T pattern [6] Voronoi diagram... system for the English and French sets 297 297 298 298 299 299 300 306 311 311 313 314 315 316 ACRONYMS 1-NN 2D 3D AC ADF ANN ANNIGMA ANSI ARG ASCII APR ARAN BAB BKS BMN BP CC CCA CD CE CI CL CRS DAG Nearest Neighbor Two-Dimensional Three-Dimensional Auto-Complexity Automatically Defined Functions Artificial Neural Network Artificial Neural Net Input Gain Measurement Approximation American National Standards... (middle), and string classes (right) 215 Segmentation candidate lattice of a word image The optimal segmentation candidate is denoted by thick lines 228 Segmentation -recognition candidate lattice of the word image of Fig 5.13 Each edge denotes a candidate pattern and its character class 228 Match the string image of Fig 5.14 with two word classes Upper: matching path in search space (array), lower:... (Paris, France), Bell Canada, Canada Post (Ottawa, Canada), DocImage Inc (Montreal, Canada), Hitachi Ltd (Tokyo, Japan), La Poste (French Post, France) Furthermore, Mohamed Cheriet is particularly grateful to his PhD student Vincent Dor´ for carefully reviewing Chapter 2 and for providing considerable assistance by e typesetting parts of a text in LaTeX Cheng-Lin Liu is grateful to his former advisors and . strings rather than elementary units of isolated characters. However, word and character string recognition processes involve character segmentation, character recognition, and con- textual analyses detection and correction of slants and skews, and the normalization of character sizes and positions. As well, methods of extracting character contours and skeletons are presented with appropriate. Data Character recognition systems : a guide for students and practioners / Mohamed Cheriet [et al.]. p. cm. Includes bibliographical references and index. ISBN 97 8-0 -4 7 1-4 157 0-1 (cloth) 1. Optical character