连体段特征聚类的维吾尔文文档图像单词切分

Connected components feature based Uyghur document image word segmentation

为有效解决文档图像中单词漏切分和过切分问题,分析维吾尔文文档图像的无嵌入式双栏版面特性和文字特点。综合考虑连体段位置信息、密度及高宽特征和相邻连体段重叠性,提出一种文档图像中精确切分完整单词块的方法。将图文混排的版面分析与重叠域合并相结合,采用两级K-means分类策略,有效避免标点符号的影响,增强完整单词块的被切分能力。实验结果表明,该算法比连通域搜索算法和投影算法具有更高的切分精度,在多文种图像单词切分中具有更高的有效性。

To deal with the problem of word over segmentation for document image,the nature of non-embedded double column document image and Uyghur text were analyzed.A kind of accurate segmentation method for whole word blocks in complex document image was proposed after considering the features such as position,density and high width,overlap of adjacent segments.Mixed layout analysis was combined with overlap domain merging,two-level K-means classification strategy was used to prevent the punctuation influences,and the word segmentation capability was enhanced.Experimental results indicate that the proposed algorithm demonstrates higher segmentation precision than the connected domain search algorithm and projection algorithm,and higher validity in word segmentation of multi-language images.