Research On Feature Extraction And Matching Recognition Of Printed Chinese Character Recognition System

Chinese character used by most people in the world with thousands years in history is different from alphabetic letters. So the research on Chinese character recognition technology is the key task in the social informatization development. In the information explosion era, how to make computers effectively "understand" so many Chinese character, especially printed Chinese character in order to save much manpower, is an important factor in Chinese character recognition technology. Improving the effect of printed Chinese character recognition, including recognizing rate and recognizing speed, has important practical and theoretical significance in OA, machine translation, AI and so on.Taking the first class totally 3755 Chinese character in the GB2312-80 as the character database, this paper elaborates Chinese character recognition technology from feature extraction and matching recognition. It does carefully research on these two aspects and makes important improvement. Compared with original algorithms, the improved algorithms perform better. The main task of the paper is as follow.(1) It takes strengths and weaknesses of both structure features and statistical features into consideration and analyzes many kinds of character features, including complexity, connected components, closed regions, coarse periphery, cellular feature, texture feature and so on. Based on the research of the feature extraction algorithms, It analyzes problems and presents certain improved algorithms that overcome or weaken the problems and provides combinatorial optimization feature with a powerful guarantee. Moreover, it also presents strokes code feature and feature points based on Chinese character strokes types, and both of them improve the recognizing effects of this Chinese character recognition system.(2) During the matching process, it elaborates usual clustering method, including ISODATA, UPGMA, K-means and one improved K-means method, and presents certain resolutions considering printed Chinese character recognition. Finally, the paper presents an improved K-means method based on optimizing initial points by UPGMA and this method effectively takes good advantages of hierarchical clustering algorithms and partitional clustering algorithms. Experiment results show that the proposed approach has higher accuracy and speed. Concentrating on the research of both Chinese character feature extraction and matching process, it takes the entire flow configuration of the system into consideration and develops a printed Chinese character recognition system and achieves satisfactory results.