Automatic System For Image Understanding Based Chinese Characters Structure Analysis

Chinese character is the only ideography which has been being used without breaking ever since it is created. Chinese character's main body is its component structure, which consists of Chinese character component structure and their combination ways. The research work on Chinese character's main body should center around Chinese character components, and study their overall combination regulation based on individual examinations, then to establish some basic theory and operational methods about Chinese character structure by analysis. By analyzing character's two-dimensional images, Chinese character component structure analysis focuses on components combination ways, components combination process and component units' topological relationship of Chinese character, aiming at reveal the regulation of Chinese character component structure.Chinese character image consists of four parts, including stroke, sub-component, component and single character. This paper generates Chinese character image using tools, and process these images by methodologies in image morphology such as thinning and 8-connected partition, to produce Chinese character sub-component images. Then, after digitalization, by extracting and classifying effective features of strokes and end-points according to the information of relations among pixels distribution, and by determining the relationship between Chinese components and their location based on predefined component knowledge library, we can recognize Chinese character component structures correctly.The automatic analysis system of Chinese character structure which simulates human recognition of Chinese character images by computers, applies many mature techniques in the fields such as computer vision, image understanding, software engineering, data mining to main parts of research work on Chinese character structure analysis. Based on subcomponent images generated from GB2312-80 Chinese character set, we have generated Chinese character components structure's feature description. And then, we have performed further research work on decomposition and recognition of Chinese character component, statistical information of Chinese character component, and automatic generation of mathematical expression of Chinese character. Also, we point out some promising directions for further study in this field.