用动态有序弹性匹配方法识别手写印刷体汉字

Recognition of Handprinted Chinese Characters with Dynamic-Sequential-Flexible matching approach

本文为手写印刷体汉字分类提供了一种新的实用解决方法。本方法采用日本山本氏提出的用松驰匹配方法识别手写印刷体汉字的基本思想。给出了用汉字轮廓点的差分作为轮廓跟踪及折断点选择的算法,提高了轮廓折线化速度;分类是按汉字四周的外轮廓线分二级进行匹配的,提出了动态有序弹性匹配方法,使全域松弛匹配得以在局部区域进行,从而有效地提高了匹配速度;在进行第一级粗分类时,采用多个模板,有效地提高了按偏旁位置分布进行分类的正确率;最后利用内轮廓折线与字典逐字匹配,达到手写印刷体汉字识别的目的。本识别系统在386微机上用C语言编程实现,对国标一级3755个汉字进行了三万余字的识别试验,前十位的累计正确分类率达92%,第一位的累计正确分类率达84%。识别时间为每字1.8秒。

This paper proposes a new practical method for classification of Handprinted ChineseCharacters, which is based on the Relaxation Matching method. It gives a algorithm of tracing thecontour and selecting the turning points by using the distances between points on the contour of chi-nese characters. Hence, the speed of extracting the turning strokes is greatly improved. The first twolevels classification matching is based on the out-contour turning strokes of chinese characters. TheRelaxation Matching in full domain is changed into the one in partial domain, which is called Dy-namic-Sequential-Flexible matching approach. As a result, the matching speed is raised efficently.In the first level classification, we adopt several models, and increase efficiently the correction rateof classifing by the positions of radicals. At last, we recognize the input character by using the inter-contour turning strokes. The Recognization System is accomplished by software (Progranning in C) on Microcomputer.A middle-scale experiment of about 30000 characters shows, with the 3755 chinese characters in-cluded in level 1 of Chinese National Standard, the correct classification rate with 10th degree candi-dates is 92% and 84% with first degree candidates. the average speed is 1.8s for one character.