利用上下文相关信息的汉字文本识别

为了改善汉字文本识别率, 本文提出了一种基于语料库统计概率的后处理方法, 该方法利用上下文相关信息, 超过词汇对于汉字文本识别, 把具有确定性边界的一个汉字序列多数情况为一个句子作为一个处理单元, 利用统计获得的字字同现概率,采用动态规划方法, 获得了令人满意的效果。

Chinese Text Recognition Using Contextual Information

In order to improve Chinese text recognition rate, in this paper we present a post processing method of corpus-based statistical probabilities. The method has used contextual information more than the lexical lever knowledge. For Chinese text recognition, a bounded seguence of Chinese characters (more often, a sentence) is processed as an unit. And the cooccurrence probability between characters and dynamic progamming strategy are employed to acquire the satisficatory recognition results.