2500个印刷体汉字识别研究

本文提出关于两级汉字识别问题的一种方法。第一级分类采用四种办法:文字边缘长度,四边编码,交叉线(对角线)编码及文字两侧负象的重心。用这四种办法每类可产生平均16个字的候选类。第一级分类大大改进了识别效率。第二级分类(识别)利用R变换算法,选出其中15个分量作为特征,按最小距离分类。对带有一定干扰的2500个黑体标准汉字样在装有硬盘的DJS-130机上进行了实验研究,正确率为98.24％,误识率为0.08％,拒识率为1.68％。

A two-stage recognition method of Chinese character is proposed in this paper. The experiment has been completed for 2500 standard Chinese characters with noise by means of computer DJS-130. The recognition rate is 98.24%.