Offline Chinese Handwriting Character Recognition through Feature Extraction

In this paper, new methods were developed to successfully identify Chinese handwriting characters. These methods are based on features extraction as compared and matched with HCL2000 database [1]. Several algorithms were applied for binarization, smoothing, noise reduction and thinning to an image of a single Chinese character. Then the image is given to a structural feature extracting algorithm, which transforms that character into an undirected graph with unique coordinates of all nodes. The resulting graph was compared with 3755 samples from the database one by one, whose features are also extracted and stored in a graph. The total deviation between two characters was obtained by comparing edges from the generated undirected graph representing a character and its counterpart graph which is generated from the image in HCL 2000 database. Based on the measurements of lengths, orientation, and areas between lines, the best match was selected as the result of recognition. Additional principles are also included in order to assure the accuracy. During the matching, the graph may sometimes be slightly transformed or modified to maximize the fitness criteria. Experimental results are accomplished by the use of 1000 random characters to test the effectiveness. The accuracy of the recognition system is significant. Analysis and experimental results are presented.