Chain-code transform for Chinese character recognition

Abstract

A new parameter transformation method, called chain-code transform (CT), has been introduced. It is particularly well-suited for the recognition of patterns containing straight lines especially handwritten Chinese characters, which consist mainly of straight strokes in four directions. The CT method essentially maps the strokes of a character into a 2-dimensional parameter space similar to that of the Hough transform. However, by the novel use of the chain-code representation of characters, the background noise problem of Hough transform is eliminated. Furthermore, the CT method has the additional advantage of quantizing the orientation axis into four directions which coincide with the main stroke directions of Chinese characters. Preliminary recognition experiments have been conducted on a small database with a total of 3600 Chinese characters written by 20 individuals. The results indicate that the CT method gives the best recognition rate when compared with projection profile method, Fourier transform, and Hough transform.