基于统计部首模型的联机手写汉字识别方法  

On-line handwritten Chinese character recognition using statistical radical models

作　　者：马龙龙[1] 刘成林[1] 

机构地区：[1]中国科学院自动化研究所模式识别国家重点实验室,北京100190

出　　处：《智能系统学报》2010年第5期385-391,共7页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(60775004;60825301)

摘　　要：利用汉字的部首层次结构有助于减小字符识别器的存储空间和提高泛化性、适应性,但部首分割一直是一个难点.提出一种新的基于部首的联机手写汉字识别方法,该方法把部首形状信息和几何信息集成到识别框架中,在组合搜索过程中利用字符-部首的层次结构字典引导部首的分割与识别,从而提高部首分割的准确率.为克服部首间的连笔,引入角点检测提取子笔划.部首识别采用统计分类器,模型参数通过自学习得到.在字符识别中,采用了2种不同的字典表示以及相应的不同搜索算法.该方法已用于左右与上下结构的字符集,实验结果表明了该方法的有效性.The hierarchical radical structure of Chinese characters can be explored to reduce the number of parame- ters in character recognition, as well as to improve the generalization ability and adaptability. However, the seg- mentation of radicals from characters has long been a difficult problem. A new radical-based approach for on-line handwritten Chinese character recognition was proposed. The approach integrated appearance-based radical recogni- tion and geometric context into a principled framework using a hierarchical character-radical dictionary to guide rad- ical segmentation and recognition during the path search process for the purpose of increasing the accuracy of radi- cal segmentation. The parameters of statistical radical models were estimated in embedded learning. To overcome the connection of strokes between radicals, corner points were detected to extract sub-strokes. For character recog- nition, two dictionary representation schemes and accordingly different search algorithms were used. The effective- ness of the proposed approach has been demonstrated on Chinese characters of left-right and up-down structures.

关 键 词：联机手写汉字识别 统计部首模型 层次结构 过分割 路径搜索 部首识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]