摘要： 字符图像的分割是票据自动处理系统中一个预处理过程，提取字符的质量对识别系统有很大的影响。我国支票大写金额图像背景中存在定位线和背景纹理，还有各式各样的印章对笔画信息的干扰。该文解决了复杂背景下二值化问题，采用形态学方法去除定位线，根据目标与背景的灰度和双边缘特征分析进行二值化处理，利用用户填写字符串时的书写习惯去除孤立印章碎块，基于区域增长的后处理过程来修补前期处理过程中断裂的笔画。该方法在真实支票图像的获取上，取得了较为满意的结果。

关键词: 阈值, 文档图像分析, 分割, 边缘

Abstract: Legal amount characters extracting from chinese bankcheck images represents an important challenge in the field of automatic check processing. Difficulties are derived mainly from the complex backgrounds, including preprinted lines, texture lines and seal images with different types and positions, since they are often dark and stroke-like. This paper proposes a new legal amount image segmentation method. Preprinted lines are removed with morphological method. The new image is pre-binaried based on the analysis of gray level feature and local double edge feature. Broken strokes are connected through region growing procedure. Comparative experiment results on real-life check images demonstrate the effectiveness of the method.

Key words: Threshold, Document image analysis, Segmentation, Edge