Knowledge-based sub-pattern segmentation: decompositions of Chinese characters

Chinese characters are complex two-dimensional patterns. Although radicals are the most significant components in Chinese characters, it is difficult to extract them from the characters because radicals are often interlaced and connected to each other and, consequently, histogram-based segmentation methods fail to separate them. We propose to segment the interlaced parts in the images by a skeletonization algorithm based on contour tracing. To separate further connected parts of the Chinese characters we suggest a new processing method resorting to the knowledge of the radical connections and the structure of the characters. The separation repeats until all the disconnected parts cannot be segmented further conforming to our knowledge-based criterion