印刷体藏文识别中字符切分方法的研究  

Research on character segmentation method in recognition of printed Tibetan

作　　者：公保杰 安见才让[1] Gong Baojie;Anjian Cairang(College of Computer Science, Qinghai Nationalities University, Xining, Qinghai 810007, China)

机构地区：[1]青海民族大学计算机学院

出　　处：《计算机时代》2019年第9期24-26,共3页Computer Era

摘　　要：印刷体藏文字符的准确切分是识别的关键,由于藏文字符结构的特殊性导致字符之间会出现重叠粘连的现象,使得切分很困难。文章提出多策略细化切分方法,首先用积分投影法实现行和单字的粗切分,再对重叠粘连的字符,根据连通域、藏文字符基线位置像素的统计、字符宽度等信息进行细切分。实验表明,该切分方法提高了印刷体藏文字符切分的准确率,为提高印刷体藏文的识别效率提供基础。The accuracy of the segmentation is the key to identify printed Tibetan characters. Due to the particularity of Tibetan character structure that characters appear overlapping adhesion phenomenon in between, makes the segmentation difficult. This paper propose a multi-strategy refined segmentation method, which uses integral projection method for a coarse segmentation to separate the lines and words, then a fine segmentation is conducted to separate the overlapping conglutination characters according to the connected domain, and the information of Tibetan character baseline position pixel statistics and the character width. Experiment shows that this segmentation method improves the accuracy of printed Tibetan character segmentation, and provides a basis for improving the printed Tibetan recognition efficiency.

关 键 词：印刷体藏文 积分投影 切分 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]