金融票据OCR系统训练样本挑选方法的研究-模式识别与智能系统专业论文.docxVIP

金融票据OCR系统训练样本挑选方法的研究 金融票据OCR系统训练样本挑选方法的研究 摘要 墒彝 J11—；■～ 本论文在国家863计划项目“金融票据OC系统中的关键技术研究”的基础 上，结合实际项目应用，比较详细地论述了该系统中手写汉字的识别部分以及手 写汉字训练部分的研究现状。对于识别部分，论文详细介绍了手写汉字识别部分 的预处理部分、特征抽取部分、识别算法原理以及基于余弦整形变换的多模板匹 配手写汉字识别方法，并且通过实际测试证实了该方法的有效性。对于系统的训 练部分，论文主要是针对训练样本挑选方法进行改进工作，提出了几种基于K 均值聚类算法的训练样本挑选方案，并且通过实际测试验证了这些方案对于提高 整个系统的性能的有效性；根据系统的需要，本文详细叙述了金融票据OCR系 统模板定义工具的创建过程。 关键词 训练样本挑选，聚类算法，K均值法，预处理，手写汉字识别，余弦整形变换 多模板匹配 北京由区电大学硕士研究生论文 一筹I页． 金融票据OCR系统训练样本挑选方法的研究 金融票据OCR系统训练样本挑选方法的研究 摘要 Abstract Based on the project of National 863 Initiative—Key Technologies of Bank Check OCR System，this paper is the research result for the application of this system．This paper discussed the detail of the hand—written Chinese character recognition subsystem and the hand-written Chinese character training subsystem of Bank Check OCR System．As for the former subsystem，the preprocessing ofthe sample images， the feature extraction of the character samples，the core recognition algorithm of the hand-written Chinese characters and a multitemplate matching hand-written Chinese character recognition algorithm based on cosine paccern transformation are discussed in details．We also tested the validity of this multitemplate roaching method through a set of recognition tests．As for the Iatter subsystem．the author summarized his research work of improving the method of training sample selection，introduced several schemes of training sample selection based on the K—Means clustering algorithm and tested their effect on the performanee of the whole system of Bank Check OCR．And according to the practical demand，the author developed a tool for the purpose of creating template definition files and discussed this process and also the usage ofthis toolin details． Keywords Training Sample Selection，Clustering Algorithm，K，Means Algorithm，Preprocessing， Recognition of Hand-written Chinese Characters，Cosine Pattern Transformation， Multitemplate Matching 北京邮电大学硕士研究