网格笔划密度特征的OPCCR错误自动检测

现有的光学印刷体汉字识别（OPCCR）系统中,汉字识别率虽然已经高达98%以上,但仍然会发生错误识别的情况.通常,这些错误的识别还不能被自动检测,采用人工检测,费时费力,大大降低了实际应用系统的自动化和智能化程度.为此,本论文提出了基于网格笔划密度特征的OPCCR错误的自动检测算法.本算法首先建立标准汉字的网格笔划密度特征的特征库.然后,在OPCCR错误的自动检测时,对光学印刷体汉字图像进行预处理、行分割、列分割得到单个汉字图像,提取单个汉字图像的网格笔划密度特征.再把特征和相应的识别出的汉字的特征库中的特征进行相关匹配.于是,根据特征匹配自动检测OPCCR的错误.

Automatic detection of OPCCR errors based on mesh grid density features of strokes

At present, although the recognition rate of optical printed Chinese character recognition (OPCCR) systems has reached high up to 98 percents, there still exist recognition errors. Normally, the errors can not be detected automatically, but manually. Hence, it wastes time and labor significantly and reduces the extent of automation and intelligence of the systems. For this reason, an algorithm of automatic detection of OPCCR errors based on mesh grid density features of strokes has been proposed in this paper. First, a database of mesh grid density features for standard Chinese characters is built up. Then, to detect OPCCR errors, the image of optical printed Chinese characters (OPCC) is preprocessed, line cut and character cut to obtain individual characters. The mesh grid density features of the individual characters are extracted. The features are matched by correlation with the mesh grid density features in the feature database of the corresponding Chinese characters recognized. Thus, the OPCCR errors are detected according to the feature matching.