小写金额的多模式切分与识别算法

针对带表格的中文支票小写金额的自动识别问题,提出了一种多模式切分和识别算法。根据小写金额不同部分的切分和识别难度,采取了3种递进的模式:预切分模式、连写0检测模式和基于识别的切分模式。其中预切分模式用来处理小写金额中不粘连的单字;连写0检测模式用来检测并识别连写的0;基于识别的切分模式用来处理非连写0的粘连部分,在这个模式中采用了遗传算法来加速最优解的搜索过程。利用从银行采集的1053张真实支票样本进行测试,在拒识率为33.6%时,小写金额串的识别率达到66.1%,实验结果证明这种算法可以提高真实支票小写金额的识别率。

Multi model Segmentation and Recognition Algorithm of Courtesy Amount

A multi-model segmentation and recognition algorithm of courtesy amount on Chinese bank checks with the form lines is presented in this paper.Based on some characteristics of Chinese bank checks,we adopt three models for different parts of the courtesy amount.The pre-segmentation model deals with the isolated characters.The touching zeros detection model is designed for the part of touching zeros.The segmentation-based recognition model deals with other touching part in the courtesy amount.In the third model,we use genetic algorithm to accelerate the searching process.The system is validated with 1 053 real bank checks.The reject rate is 33.6%,and the recognition rate at the amount level can reach 66.1%.The experiment results show that the recognition rate of the real bank checks can be improved with this new method.