Substring Alignment Method for Lexicon Based Handwritten Chinese String Recognition and Its Application to Address Line Recognition

This paper presents a lexicon based method for Chinese string recognition. In this method, we recognize a Chinese string image as a whole by matching it against lexicons in a database. We first over-segment the input line image into a series of radicals and recognize all the possible radical combinations. We then search for candidate lexicons in a given database according to the extracted keywords. Each lexicon is compared with the image to find the best match between the radicals and the given string by substring alignment. In this process, the segmentation and recognition results are determined synchronously. Our method is tested on 500 handwritten images of Chinese address and achieves a correct rate of 87% in address match