银行支票中小写金额图像的提取

支票图像的分割与识别是目前文档自动处理领域中讨论的一个热点问题。其中字符图像的分割是预处理过程中的一个重要环节,对识别系统有很大的影响。在我国的支票图像中常含有较深的印章图像叠加在被分割的字符上,增加了字符分割的难度。本文以支票中小写金额的图像为例,提出了字符的逐层分割方法以及用于判断印章图像是否去除的评判准则。首先去除图像中的底纹和定位格线;然后通过迭代的方法选取阈值去除印章图像;最后采用基于连通区的区域增长算法提取字符图像,去除碎块。在2725张实地采集的我国现行支票上的实验结果表明,本文的方法能够有效的去除印章图像,分割出字符。

Extraction of Courtesy Amount Image in Bankcheck

Characters extracting from check images represents an important challenge in the field of bankcheck processing system.Difficulties in segmentation of the Chinese check images are derived mainly from the different types and positions of the seal imprints,which is often dark and stroke like.In this paper,a rule is proposed to evaluate the segment results of the Chinese check images,and a new segment method based on it is also developed.Background and preprint lines are first removed;then seals are segmented by threshold determined by proposed rule;finally,region growing process is used to remove noise.Experiments on 2725 real lift bank checks demonstrate the effectiveness of the proposed method as compared with four other common used segment methods.