Research On Off-line Handwritten Chinese Characters Recognition Based On The Binary Tree Of SVM Multi-classification

Chinese character recognition has been one of the most important area of research in the pattern recognition,has broad application prospects.After years of research,has made a lot of results. However, the non-binding non-specific off-line handwritten Chinese character recognition is still considered as one of the most difficult problem in character recognition area, the reason can be summed up as follows: more similar to Chinese characters, and some are very subtle differences between similar words; there is a large number of irregular writing deformation.SVM (Support Vector Machine) has been in pattern recognition, regression analysis and feature selection and other aspects of good results obtained. Therefore, based on support vector machine off-line handwritten Chinese character recognition research has a certain theoretical significance and application value.Off-line handwritten Chinese character for the study object,this paper givesa judge standard of the complexity of Chinese characters to do initial coarse classification of Chinese characters. Then based on the font structure and type of the external border the paper do further coarse classification of Chinese characters, and finally divided into nine kinds of character types, generates a binary tree.Accroding to the character set type of binary tree leaf node, choose a different mix of features as the SVM classifier of each group's input, using "one to one" approach to accomplish the final fine classification. This paper made the following specific research areas:(1) The structure of coarse classification of the binary tree of Chinese characters. Accroding to the relationship between handwritten Chinese character stroke number and complexity to do initial coarse classification;then do further coarse classification of Chinese characters sets by studying the characteristics of the font structure and the external border of handwritten Chinese characters, to construct a coarse classification of handwritten Chinese characters binary tree, implemented multi-level coarse classification based on the complexity, font structure, border type of Chinese characters. (2) The improvement of SVM kernel function parameters optimization met-hod. A optimization method of kernel parameters was improved based on kernel-target alignment theory, use improved method to cluster on the training samples before training to obtain the kernel parameters.(3) The research of feature extraction and fusion method of off-line handw-ritten Chinese character. Based on the different characteristics of different Chinese character sets in coarse classification of the binary tree, use of different features as the input of various SVM classifier; similarly, based on different Chinese character sets of different characteristics, select the integration of differ-ent features and the formation of its new features, as each "one to one" SVM classifier input for fine classification.(4) In the simulation,the handwritten Chinese characters in SCUT-IRAC are chosen and make the MATLAB7.0 as simulation tool.The results of simulation showed that: In this paper, the binary tree SVM coarse classification and "one to one" SVM fine classification new method of combining classification and recognition give full play of the binary tree classification speed and SVM classification accuracy advantage of the high rate, there are good results.