动态兼容线性分类法识别印刷体汉字

Dynamic Compatible Linear Classification for Recognition of Printed Chinese Characters

本文针对目前汉字识别中两个关键问题——特征的选择和识别方法进行了较深入的研究。提出了利用动态兼容线性分类法实现一个印刷体汉字识别实验系统。在分类中,采用了质量等十一种特征。其中四象限投影脉冲恃征,八象限特征,十六象限0—1特征和四方向微分特征是本文提出和修正改进的。本文还提出了动态兼容线性分类法和识别字典自动完善的方法。利用上述方法在微机上构成了一个实验识别系统。对3000个加有随机噪声的标准汉字进行了识别,经过训练后识别率可达98.5％,识别速度平均每字0.4秒,收到较好效果。

In this paper, we have made a deep study to two key problems at present in Chinese character recognition-choice of characteristic and method of recognition A trial system using dynamic compatible linear classification to complete a printing Chinese character recognition is put forward. 11 kinds of characteristic, such as quality are adopted in classification Among those characeeristic of projective pulse in quadrant 4, characteristic in quadrant 8,0-1 characteristic in quadrant 16 and differential characteristic of 4 directions have been presented and improved in the paper At the same time, a method of dynamic compatible linear classification and a method of auto natically improved recognition dictionary also have been introduced. A trial discernmemt system using above method is constituted with the microcomputer and 3000 standard Chinese characters that carried random noise are discerned After training the rate of recognition is 98.5%, the speed of recognition on the average is 0.4 second per Chinese character and the result at last is better.