基于神经网络和模糊匹配算法的手写汉字预分类研究  

Study on preclassification for handwritten Chinese character based on neural net and fuzzy matching algorithm

作　　者：卢达[1] 浦炜[1] 陈琦玮[1] 谢铭培[2] 

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

出　　处：《计算机应用》2005年第10期2418-2421,共4页journal of Computer Applications

基　　金：江苏省教育厅自然科学基金资助项目(02KJD540001)

摘　　要：对手写汉字识别问题,提出了一种在识别之前对手写汉字预分类的新方法,该方法用Neocognitron网提取字符笔画特征,然后采用有监督的扩展ART神经网络(SEART)产生一定数量的预分类组并通过基于模糊相似测量的匹配算法进行预分类。实验表明,该方法用于手写汉字分类效果良好,预分类正确率达到98.22%。To settle the recognition task of handwritten Chinese characters, the authors put forward a method for handwritten Chinese character preclassification before character recognition. In this method, Neocognitron was used in extracting stroke features, then uses the Supervised Extended ART （SEART） to create some preclassification groups, and uses matching algorithm of fuzzy prototypes of similarity measurement for character preclassification. The experiment shows this method is effective when used for handwritten Chinese character classification and characters of the testing set can be distributed into correct preclassification classes at a rate of 98.22%.

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

分 类 号：TP391[自动化与计算机技术—计算机应用技术]