A knowledge-based stroke-matching method for Chinese character recognition

Due to the complexity and variations of handprinted Chinese characters, elaborate structural-matching algorithms are required for character recognition. Although the iterative stroke-matching method proposed by Yamamoto and Rosenfeld is very powerful, it is very computationally intensive. Moreover, the character set used by the Chinese is greater in both si/e and complexity than that used by the Japanese, and experiments performed by the authors indicate that the algorithm is too computationally intensive to be implemented for the recognition of complicated Chinese characters. A modified relaxation method that greatly reduces the computational load is proposed. The idea is to incorporate specific knowledge about the Chinese characters into the training system to remove inconsistencies with a higher efficiency and to speed up the matching process. Experimental results on the 240 most frequently used characters written by 37 different individuals showed that a four-fold improvement in speed could be achieved, making it more feasible to implement the algorithm for the recognition of complicated Chinese characters.