The Research Of Large-Set Offline Handwritten Chinese Characters Recognition

The offline handwritten Chinese characters recognition is popular in the region of Chinese characters recognition. Handwritten Chinese characters are characterized with large vocabulary, complex structure, lots of similar characters and serious irregular variations of shapes. In this paper, we take a preliminary research on the offline handwritten Chinese characters recognition based on the precursors' work and achievement in this field, and get some useful conclusion. Main work is as follows:1. Greatly refined the experiment system built by precursors greatly, and carried on a great deal of comparison experiments taking the HCL2000 handwritten Chinese characters database as the foundation.2. Studied the influence which character features have on the classification rate, and the performance of the combined features.3. Studied how various distance classifiers influence on the Chinese characters features. Proposed a method to improve the performance of fine classifier based on multiple classifiers combination.4. We designed a realization solution using one-class-one-set scheme to carry on character fine classification, and got a fine recognition ratio at 86.6%.5. Gave an introduction about tangent distance and tangent vector. We managed to take tangent distance with SVD method as fine classifier. We studied how the selection of tangent vector's number influences the performance of the classifier. We got a high fine recognition ratio at 99.2% when we selected 8*8 meshed contour direction feature as fine classification feature, 30 tangent vectors, and proved the efficiency of the tangent distance classifier.6. Studied how the different size of coarse candidate set influence the Chinese character recognition system's performance, and gave a conclusion on how to select the size of the candidate set.In the finality, the problems requiring further studies are discussed.