维汉英混排文档识别

Uyghur, Chinese and English Multilingual Document Recognition

维、汉、英是特点完全不同的文字。该文依据多层次语言判断和适当干预的多语言字符识别系统设计原则首次实现了维、汉、英混排文本识别系统。识别系统首先根据维、汉、英文字的各自特点实现字符块语言属性的初步判断,然后针对每种文字设计不同的字符切割算法。字符识别可信度用来判断字符语言属性和字符切分结果是否正确。实验结果表明,各种维、汉、英混排文本识别率达到96.4%以上。

The characteristics of Uyghur, Chinese and English scripts are totally different. A Uyghur, Chinese and English multilingual document recognition system is implemented the first time based on the multilingual OCR system design principle, which includes ＂multi-layer character language estimation＂ and ＂suitable adjustment＂. At first, the language property of each text block is estimated according to the characteristics of Uyghur, Chinese and English scripts. After that, language-oriented character segmentation algorithms are performed on text blocks, and the character recognition confidence is used to judge whether the results of character segmentation and language property estimation of a text block are right. Experimental results show the recognition accuracy of Uyghur, Chinese and English multilingual documents achieves 96.4% and above.