基于SEART网和模糊相似测量的手写汉字预分类法  

Preclassification for handwritten Chinese character based on SEART neural-net and fuzzy similarity measure

作　　者：卢达[1] 浦炜[1] 谢铭培[2] 

机构地区：[1]常熟理工学院物理与电子科学系,常熟215500 [2]上海复旦大学计算机科学系,上海200433

出　　处：《东南大学学报（自然科学版）》2005年第A02期79-83,共5页Journal of Southeast University：Natural Science Edition

基　　金：江苏省教育厅自然科学研究资助项目(02KJD540001)

摘　　要：提出了一种手写汉字预分类的新方法.该方法分2步进行,首先采用有监督的扩展ART神经网络(SEART)产生一定数量的预分类组,然后通过模糊逻辑处理将各组字符分别转换成基于非线性加权函数的模糊样板,并通过基于模糊相似测量的匹配算法、相似性测量样板的分级分类进行预分类.测试结果表明,该方法效果良好,预分类正确率达到98.19%.A method of character preclassification for handwritten Chinese character recognition is proposed. Two stages are employed： in stage Ⅰ, the supervised extended ART （SEART） is used to create some preclassi- fication groups; in stage Ⅱ, the characters in each group is transformed into fuzzy prototypes based on a nonlinear weighted similarity function by fuzzy logic approach, then the matching algorithm and hierachic classification of fuzzy prototypes of similarity measurement for character preclassification are used. The experimental result shows that this method is effective and the characters of the testing set can be distributed into correct preclassification classes at a rate of 98.19%.

关 键 词：手写汉字预分类 人工神经网络 有监督的扩展ART 模糊相似测量 匹配算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]