Research On Fusion Features Of Free Off-line Handwritten Chinese Character Recognition

In the field of pattern recognition, pattern deformation is everywhere, how to deal with the adverse effects of pattern deformation to feature extraction and late recognition, which is an important issue to be solved urgently. In the field of Chinese character recognition, during the writing of Chinese characters, Chinese characters will exist many deformations. And the research works on the severe deformation of cursive, running scripts and writing style of some professional field personnel are relatively less. This paper will take the large deformation of handwritten Chinese characters as the research object, and a series of research on the features of free handwritten Chinese characters have been done. The study on the features of free handwritten Chinese characters will provide new ideas and methods for the study of the deformation of other modes. In summary, the study of free offline handwritten Chinese character recognition feature has great significance of further development in the field of pattern recognition.Feature extraction is one of the most critical aspects of off-line handwritten Chinese character recognition, and how to extract the excellent features of Chinese character is one direction that many scholars have devoted to studying. The concrete research content is as follows:(1) According to the content of this subject, collection program and collection process of free handwritten Chinese character sample has been planned; the establishment of free Character sample library also has been set up. Meanwhile, each sample of the library of free handwritten Chinese characters has been processing, including gray-scale, binarization and normalization.(2) In this paper, a new feature fusion method has been proposed, which is based on image multi-resolution techniques. At the same time, parallel feature fusion method,which is based on the proposed method, is applied to off-line handwritten Chinese characters recognition in large deformation. And optimizing selection of the combination coefficient of parallel features and the new feature dimension d in K-L transformation are made to obtain relatively stable recognition results.(3) The different methods of feature extraction and feature fusion schemes are carried on the contrast analysis, which finally validate that the proposed method is effective; meanwhile, method presented in this paper with others in the literature methods were compares to analysis, which verify the feasibility and effectiveness of the method proposed in this paper furthermore.