基于字符边缘特征的手写账号切分算法  

Segmentation Algorithm for Unconstrained Handwritten Account Numbers Based on Edge Feature

作　　者：张闯[1] 吴铭[1] 郭军[1] 

机构地区：[1]北京邮电大学模式识别与智能信息系统实验室,北京100876

出　　处：《计算机工程》2003年第21期34-35,72,共3页Computer Engineering

基　　金：国家"863"计划基金资助项目(2001AA114080)

摘　　要：结合银行票据自动识别系统（Bank-OCR）的开发研究，提出了基于手写数字串边 缘特征的手写数字串的切分算法。文章分析了手写数字串的切分难点，提出了数字账号边缘 特征的概念，并在利用下边缘特征判断连体字符个数的基础上，利用上下边缘中波峰波谷的 特征来对连写数字进行有效切分。实验结果表明对于非限制自由手写数字串的最终切分正切 率达到 86.3%，满足了银行票据的预处理要求。This thesis focuses on unconstrained handwritten account numbers seg mentation algorithm based on edge feature, which is most important for the devel opment of the bank check OCR system. The paper analyzes the difficulties of the unconstrained handwritten account numbers segmentation, and proposes the concept of the edge feature and the length estimation of connected components using the feature of lower edge also be discussed. The segmentation algorithm based on th e feature of the wave crest and trough on the upper and lower edge is proposed. This segmentation algorithm that has promising results (the segmentation correct rate reach to 86.3%) and is then incorporated into a complete bank check charac ter recognition system.

关 键 词：票据处理 手写账号 字符切分 边缘特征 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]