A Handwritten Chinese Address Recognition Method Using Recursive Holistic Word Matching and Edit Distance Based Verification Strategies
Kaizhu Huang, Yoshinobu Hotta, Satoshi Naoi · 2006
With extensive applications in many domains, e.g., in postal address recognition, bank cheque recognition etc, Hand-written Chinese Address Recognition (HCAR) has attracted much interest recently. However, due to the large num-ber in Chinese characters categories, irregular distortion and touching in handwriting, and high variations in writ-ing styles, this task presents a big challenge in the com-munity of pattern recognition. In this paper, we propose a novel three-stage approach to attack the Handwritten Chi-nese Address Recognition problem. More specifically, we propose the framework of coarse recognition, fine recogni-tion, and verification as well as some new techiqniues tai-lored to each stage for HCAR. Coarse recognition is de-signed for speeding up the system; fine recognition exploits a novel recursive and enhanced segmentation-free holisti-cal word recognition method in order to overcome the dif-ficulties in HCAR; verification uses an important modified edit distance definition for further lifting the system accu-racy. A series of experiments are conducted to evaluate the proposed novel system. The results demonstrate that our system outperforms other traditional methods in three real data sets. 1