لایب سان

Organization of the Monograph Optical character recognition (OCR) is one of the most popular areas of research in pattern recognition [3, 25] since past few decades. It is an actively studied topic in industry and academia [8, 15, 18, 24] because of its immense application poten- tial. OCR was initially studied in early 1930s [23]. It has its origins in Germany as a patent by Gustav Tauschek. OCR is a technique of translating handwritten, type- written or printed text characters to a machine-encoded text [23, 25]. It is a pro- cess of reading handwritten characters and recognizing them. It is widely used as a form of data entry from printed paper data records which may include passport documents, invoices, bank statements, computerized receipts, business cards, mail, printouts etc. OCR is also recognized as the subdomain of image processing which is an important research area of pattern recognition. The human brain generally finds some sort of relation predominantly in graphical form in order to remember it and recognize later. In a way it tends to produce or find patterns in handwrit- ten characters. This led to the major motivation towards the development of OCR systems. The characters of various available languages are based on the lines and curves. An OCR can be easily designed to recognize them

Contentsx 6.8.3 Hierarchical Fuzzy Bidirectional Recurrent Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 6.9 Further Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 7 Optical Character Recognition Systems for Latin Language . . . . . . . 165 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 7.2 Latin Language Script and Experimental Dataset . . . . . . . . . . . . . 167 7.3 Challenges of Optical Character Recognition Systems for Latin Language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 7.4 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 7.5 Data Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 7.5.1 Text Region Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 7.5.2 Skew Detection and Correction . . . . . . . . . . . . . . . . . . . . . . . 171 7.5.3 Binarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172 7.5.4 Noise Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 7.5.5 Character Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 7.5.6 Thinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 7.6 Feature Selection Through Genetic Algorithms . . . . . . . . . . . . . . . 175 7.7 Feature Based Classification: Sate of Art . . . . . . . . . . . . . . . . . . . . 178 7.7.1 Feature Based Classification Through Rough Fuzzy Multilayer Perceptron. . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 7.7.2 Feature Based Classification Through Fuzzy and Fuzzy Rough Support Vector Machines . . . . . . . . . . 179 7.7.3 Feature Based Classification Through Hierarchical Fuzzy Rough Bidirectional Recurrent Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 7.8 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 7.8.1 Rough Fuzzy Multilayer Perceptron . . . . . . . . . . . . . . . . 180 7.8.2 Fuzzy and Fuzzy Rough Support Vector Machines . . . . . 183 7.8.3 Hierarchical Fuzzy Rough Bidirectional Recurrent Neural Networks . . . . . . . . . . . . . . . . . . . . . . . 186 7.9 Further Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 8 Optical Character Recognition Systems for Hindi Language. . . . . . . 193 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 8.2 Hindi Language Script and Experimental Dataset . . . . . . . . . . . . . 196 8.3 Challenges of Optical Character Recognition Systems for Hindi Language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 8.4 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 8.5 Data Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 8.5.1 Binarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 8.5.2 Noise Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 8.5.3 Skew Detection and Correction . . . . . . . . . . . . . . . . . . . . 201 8.5.4 Character Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . 201 8.5.5 Thinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 Contents xi 8.6 Feature Extraction Through Hough Transform . . . . . . . . . . . . . . . 202 8.7 Feature Based Classification: Sate of Art . . . . . . . . . . . . . . . . . . . . 204 8.7.1 Feature Based Classification Through Rough Fuzzy Multilayer Perceptron. . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 8.7.2 Feature Based Classification Through Fuzzy and Fuzzy Rough Support Vector Machines . . . . . . . . . . 205 8.7.3 Feature Based Classification Through Fuzzy Markov Random Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 8.8 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 8.8.1 Rough Fuzzy Multilayer Perceptron . . . . . . . . . . . . . . . . 206 8.8.2 Fuzzy and Fuzzy Rough Support Vector Machines . . . . . 208 8.8.3 Fuzzy Markov Random Fields . . . . . . . . . . . . . . . . . . . . . 208 8.9 Further Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 9 Optical Character Recognition Systems for Gujrati Language . . . . . 217 9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 9.2 Gujrati Language Script and Experimental Dataset . . . . . . . . . . . . 219 9.3 Challenges of Optical Character Recognition Systems for Gujrati Language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220 9.4 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 9.5 Data Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 9.5.1 Binarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 9.5.2 Noise Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 9.5.3 Skew Detection and Correction . . . . . . . . . . . . . . . . . . . . 225 9.5.4 Character Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . 225 9.5.5 Thinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 9.6 Feature Selection Through Genetic Algorithms . . . . . . . . . . . . . . . 226 9.7 Feature Based Classification: Sate of Art . . . . . . . . . . . . . . . . . . . . 228 9.7.1 Feature Based Classification Through Rough Fuzzy Multilayer Perceptron. . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 9.7.2 Feature Based Classification Through Fuzzy and Fuzzy Rough Support Vector Machines . . . . . . . . . . 230 9.7.3 Feature Based Classification Through Fuzzy Markov Random Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230 9.8 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 9.8.1 Rough Fuzzy Multilayer Perceptron . . . . . . . . . . . . . . . . 231 9.8.2 Fuzzy and Fuzzy Rough Support Vector Machines . . . . . 231 9.8.3 Fuzzy Markov Random Fields . . . . . . . . . . . . . . . . . . . . . 235 9.9 Further Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 10 Summary and Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 10.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 10.2 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 xiii List of Figures Figure 1.1 The chapter wise distribution of the OCR systems in the monograph. (OCR system 1: Fuzzy Multilayer Perceptron [8]; OCR system 2: Rough Fuzzy Multilayer Perceptron [9, 10, 20]; OCR system 3: Fuzzy and Fuzzy Rough Support Vector Machines [5, 6, 7]; OCR system 4: Hierarchical Fuzzy Bidirectional Recurrent Neural Networks [4]; OCR system 5: Fuzzy Markov Random Fields [28]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Figure 1.2 The MATLAB character classifier graphical user interface . . . . 6 Figure 2.1 The different areas of character recognition . . . . . . . . . . . . . . . 12 Figure 2.2 OCR–A font . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Figure 2.3 OCR–B font . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Figure 2.4 The components of an OCR system . . . . . . . . . . . . . . . . . . . . . . 16 Figure 2.5 The baseline extraction using attractive and repulsive network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Figure 2.6 a Slant angle estimation near vertical elements. b Slant angle estimation average slant angle . . . . . . . . . . . . . . . 20 Figure 2.7 The normalization of characters . . . . . . . . . . . . . . . . . . . . . . . . . 21 Figure 2.8 The contour direction and bending point features with zoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Figure 2.9 The topological features: Maxima and minima on the exterior and interior contours, reference lines, ascenders and descenders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Figure 2.10 Sample arabic character and the chain codes of its skeleton. . . . . 27 Figure 2.11 a The deformable templates: deformations of a sample digit. b The deformable templates: deformed template superimposed on target image with dissimilarity measures . . . . 30 Figure 3.1 The computational constituents in soft computing . . . . . . . . . . 44 Figure 3.2 The variable temperature represented through various degrees of fuzziness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Figure 3.3 A multi-layer ANN interconnected through a group of nodes . . . 50 List of Figuresxiv Figure 3.4 The genetic algorithm evolution flow . . . . . . . . . . . . . . . . . . . . 52 Figure 3.5 The generation wise evolution in genetic algorithm . . . . . . . . . 52 Figure 3.6 The feature selection process using genetic algorithm . . . . . . . 58 Figure 3.7 Intra and inter-module links . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 Figure 3.8 Steps for designing a sample modular RFMLP . . . . . . . . . . . . . 64 Figure 3.9 Chromosomal representation . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Figure 3.10 The separating hyperplane between classes leading to different support vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 Figure 3.11 The graphic representation of hyperbolic tangent kernel for real values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 Figure 3.12 Long short term memory block with one cell . . . . . . . . . . . . . . 74 Figure 3.13 The trapezoidal membership function defined by trapezoid (x; a, b, c, d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 Figure 3.14 The architecture of proposed HBRNN model . . . . . . . . . . . . . . 77 Figure 3.15 Three-dimensional type-2 fuzzy membership function a primary membership with (thick dashed line) lower and (thick solid line) upper membership functions where h(o) and ̄h(o) are lower and upper bounds of the primary membership of the observation o; the shaded region is the foot print of uncertainty b gaussian secondary membership function c interval secondary membership function d mean μ has a uniform membership function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 Figure 4.1 OCR–A font . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 Figure 4.2 OCR–B font . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 Figure 4.3 A sample text from IAM database . . . . . . . . . . . . . . . . . . . . . . . 88 Figure 4.4 An image before and after thinning . . . . . . . . . . . . . . . . . . . . . . 92 Figure 4.5 The disproportionate symmetry in characters ‘p’ and ‘t’ . . . . . . 98 Figure 4.6 The character ‘t’ with slant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Figure 4.7 The uneven pixel combination for character ‘k’ (region-i on the left image has less white space than the right image) . . . . . 98 Figure 4.8 The test image for small letters . . . . . . . . . . . . . . . . . . . . . . . . . 101 Figure 4.9 The test image for capital letters . . . . . . . . . . . . . . . . . . . . . . . . 102 Figure 4.10 The comparative performance of soft computing versus traditional techniques for English language . . . . . . . . . . . . . . . . 105 Figure 5.1 ISO/IEC 8859 font . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 Figure 5.2 A sample text snapshot from IRESTE IRONFF database . . . . . 112 Figure 5.3 The camera captured image and the text regions extracted from it . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 Figure 5.4 The calculation of skew angle from bottom profile of a text region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 Figure 5.5 a The horizontal histogram of text regions for their segmentation (skewed text region and its horizontal histogram). b The horizontal histogram of text regions List of Figures xv for their segmentation (skew corrected text region and its horizontal histogram) . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Figure 5.6 a Skew correction and segmentation of text regions (an extracted text region). b Skew correction and segmentation of text regions (characters segmented from de-skewed text region). . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Figure 5.7 An image before and after thinning . . . . . . . . . . . . . . . . . . . . . . 120 Figure 5.8 The disproportionate symmetry in characters ‘þ’ and ‘ï’ . . . . . . 126 Figure 5.9 The character ‘ï’ with slant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 Figure 5.10 The uneven pixel combination for character ‘È’ (region-i on the left image has less white space than the right image) . . . . . 127 Figure 5.11 The test image for small letters . . . . . . . . . . . . . . . . . . . . . . . . . 127 Figure 5.12 The test image for capital letters . . . . . . . . . . . . . . . . . . . . . . . . 130 Figure 5.13 The comparative performance of soft computing versus traditional techniques for French language . . . . . . . . . . . . . . . . 132 Figure 6.1 A sample text snapshot from InftyCDB-2 database . . . . . . . . . . 140 Figure 6.2 The camera captured image and the text regions extracted from it . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Figure 6.3 The calculation of skew angle from bottom profile of a text region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 Figure 6.4 a The horizontal histogram of text regions for their segmentation (skewed text region and its horizontal histogram). b The horizontal histogram of text regions for their segmentation (skew corrected text region and its horizontal histogram) . . . . . . . . . . . . . . . . . . 146 Figure 6.5 a Skew correction and segmentation of text regions (an extracted text region). b Skew correction and segmentation of text regions (characters segmented from de-skewed text region). . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 Figure 6.6 An image before and after thinning . . . . . . . . . . . . . . . . . . . . . . 148 Figure 6.7 The feature selection process using genetic algorithm . . . . . . . 150 Figure 6.8 The disproportionate symmetry in characters ‘p’ and ‘t’ . . . . . 155 Figure 6.9 The character ‘t’ with slant . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 Figure 6.10 The uneven pixel combination for character ‘k’ (region-i on the left image has less white space than the right image) . . . . 155 Figure 6.11 The test image for small letters . . . . . . . . . . . . . . . . . . . . . . . . . 156 Figure 6.12 The test image for capital letters . . . . . . . . . . . . . . . . . . . . . . . . 158 Figure 6.13 The comparative performance of soft computing versus traditional techniques for German language . . . . . . . . . . . . . . . 160 Figure 7.1 A sample text snapshot from ISO Latin-1 database . . . . . . . . . . 168 Figure 7.2 The camera captured image and the text regions extracted from it . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Figure 7.3 The calculation of skew angle from bottom profile of a text region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172 List of Figuresxvi Figure 7.4 a The horizontal histogram of text regions for their segmentation (skewed text region and its horizontal histogram). b The horizontal histogram of text regions for their segmentation (skew corrected text region and its horizontal histogram) . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 Figure 7.5 a Skew correction and segmentation of text regions (an extracted text region). b Skew correction and segmentation of text regions (characters segmented from de-skewed text region). . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Figure 7.6 An image before and after thinning . . . . . . . . . . . . . . . . . . . . . . 175 Figure 7.7 The feature selection process using genetic algorithm . . . . . . . 178 Figure 7.8 The disproportionate symmetry in characters ‘þ’ and ‘ï’ . . . . . . 182 Figure 7.9 The character ‘ï’ with slant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 Figure 7.10 The uneven pixel combination for character ‘È’ (region-i on the left image has less white space than the right image) . . . . 183 Figure 7.11 The test image for small letters . . . . . . . . . . . . . . . . . . . . . . . . . 183 Figure 7.12 The test image for capital letters . . . . . . . . . . . . . . . . . . . . . . . . 185 Figure 7.13 The comparative performance of soft computing versus traditional techniques for Latin language. . . . . . . . . . . . . . . . . . 188 Figure 8.1 The ISCII code chart IS 13194:1991 . . . . . . . . . . . . . . . . . . . . . 194 Figure 8.2 The Hindi language consisting of 11 vowels and 33 consonants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 Figure 8.3 The HP labs India Indic handwriting Hindi (Devanagari) dataset. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 Figure 8.4 An image before and after thinning . . . . . . . . . . . . . . . . . . . . . . 202 Figure 8.5 The disproportionate symmetry in characters ‘ए’ and ‘इ’ . . . . . 209 Figure 8.6 The character ‘ए’ with slant . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Figure 8.7 The uneven pixel combination for character ‘क’ (region-i on the left image has less white space than the right image) . . . . . 209 Figure 8.8 The test image for Hindi letters . . . . . . . . . . . . . . . . . . . . . . . . . 210 Figure 8.9 The comparative performance of soft computing versus traditional techniques for Hindi language . . . . . . . . . . . . 212 Figure 9.1 The ISCII code chart for Gujrati characters . . . . . . . . . . . . . . . . 218 Figure 9.2 The Gujrati language consisting of 28 vowels and 36 consonants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Figure 9.3 The Gujrati character recognition dataset . . . . . . . . . . . . . . . . . 222 Figure 9.4 An image before and after thinning . . . . . . . . . . . . . . . . . . . . . . 226 Figure 9.5 The feature selection process using genetic algorithm . . . . . . . 229 Figure 9.6 The disproportionate symmetry in characters ‘ ’ and ‘ ’ . . . . . 232 Figure 9.7 The character ‘ ’ with slant . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Figure 9.8 The uneven pixel combination for character ‘ ’; region-i on the left image has less white space than the right image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Figure 9.9 The test image subset for Gujrati letters . . . . . . . . . . . . . . . . . . 233 Figure 9.10 The comparative performance of soft computing versus traditional techniques for Gujrati language . . . . . . . . . . . . . . . . 237 xvii List of Tables Table 3.1 Fuzzy set membership functions defined on Hough transform accumulator cells for line detection (x and y denote height and width of each character pattern) . . . . . . . . . . . . . . . . . . . . . 56 Table 4.1 The experimental results for the English characters using FMLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Table 4.2 The experimental results for the English characters using RFMLP. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Table 4.3 The training set results corresponding to small letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 Table 4.4 The training set results corresponding to small letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Table 4.5 The training set results corresponding to capital letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Table 4.6 The training set results corresponding to capital letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 Table 4.7 The training set results corresponding to both small and capital letters (for FSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . 104 Table 4.8 The training set results corresponding to both small and capital letters (for FRSVM). . . . . . . . . . . . . . . . . . . . . . . . . 105 Table 5.1 Fuzzy set membership functions defined on Hough transform accumulator cells for line detection (x and y denote height and width of each character pattern) . . . . . . . . . . 121 Table 5.2 The experimental results for a subset of the French characters using RFMLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Table 5.3 The training set results corresponding to small letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 Table 5.4 The training set results corresponding to small letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 Table 5.5 The training set results corresponding to capital letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 List of Tablesxviii Table 5.6 The training set results corresponding to capital letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Table 5.7 The training set results corresponding to both small and capital letters (for FSVM) . . . . . . . . . . . . . . . . . . . . . 131 Table 5.8 The training set results corresponding to both small and capital letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . 132 Table 5.9 The experimental results for a subset of the French characters using HFBRNN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 Table 6.1 The experimental results for a subset of the German characters using RFMLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Table 6.2 The training set results corresponding to small letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 Table 6.3 The training set results corresponding to small letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Table 6.4 The training set results corresponding to capital letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 Table 6.5 The training set results corresponding to capital letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Table 6.6 The training set results corresponding to both small and capital letters (for FSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Table 6.7 The training set results corresponding to both small and capital letters (for FRSVM). . . . . . . . . . . . . . . . . . . . . . . . . 160 Table 6.8 The experimental results for a subset of the German characters using HFBRNN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Table 7.1 The experimental results for a subset of the Latin characters using RFMLP Latin language . . . . . . . . . . . . . . . . . . 181 Table 7.2 The training set results corresponding to small letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 Table 7.3 The training set results corresponding to small letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 Table 7.4 The training set results corresponding to capital letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 Table 7.5 The training set results corresponding to capital letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Table 7.6 The training set results corresponding to both small and capital letters (for FSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Table 7.7 The training set results corresponding to both small and capital letters (for FRSVM). . . . . . . . . . . . . . . . . . . . . . . . . 188 Table 7.8 The experimental results for a subset of the Latin characters using HFBRNN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 Table 8.1 Fuzzy set membership functions defined on Hough transform accumulator cells for line detection (x and y denote height and width of each character pattern) . . . . . . . . . . 203 Table 8.2 The experimental results for a subset of the Hindi characters using RFMLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 List of Tables xix Table 8.3 The training set results corresponding to Hindi letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 Table 8.4 The training set results corresponding to Hindi letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 Table 8.5 The training set results corresponding to Hindi letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 Table 8.6 The training set results corresponding to Hindi letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 Table 8.7 The experimental results for a subset of the Hindi characters using FMRFs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 Table 9.1 The experimental results for a subset of the Gujrati characters using RFMLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 Table 9.2 The training set results corresponding to Gujrati letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234 Table 9.3 The training set results corresponding to Gujrati letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Table 9.4 The training set results corresponding to Gujrati letters (for FSVM). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 Table 9.5 The training set results corresponding to Gujrati letters (for FRSVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 Table 9.6 The experimental results for a subset of the Gujrati characters using FMRFs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

برای ارسال نظر لطفا وارد شوید یا ثبت نام کنید