Abstract: Complex background including shading, seal and some images has a bad effect on character recognition of check. Thus, in this paper, an effective method that extracts the significant texture characteristics of character from the image by fast lifting wavelet transform was proposed to solve this problem. A coarse-to-fine searching strategy was adopted to distinguish the characters from background at the level of block of pixels and single pixel. First, Support Vector Machine (SVM) was used to classify blocks according to texture characteristics, and during this process text region could be located. Then character segmentation was achieved by using K-means algorithm for clustering the pixels at the text region. The experimental results show the high accuracy and strong robustness of the proposed method at the situation of strong interference of complex texture and seal.

Key words: character segmentation, fast lifting wavelet transform, Support Vector Machine (SVM), K-means clustering

摘要： 针对票据字符识别中图像存在的底纹、印章和图案等复杂背景干扰问题，提出一种有效的字符分割方法。通过快速提升小波变换提取出图像中具有显著性的字符纹理特征。采用一种由粗到精的搜索策略，在图像区域和像素两个层次上逐步区分出文字和背景。首先根据区域纹理特征，利用支持向量机对区域进行分类，定位出包含文字的图像区域;然后采用K-means算法对文字区域内的像素进行聚类划分，从而实现文字分割。实验结果表明，方法具有较高的准确性，并且在背景纹理和印章干扰的情况下具有较好的鲁棒性。

关键词: 字符分割, 快速提升小波变换, 支持向量机, K-means聚类