【作者】 张捷；

【导师】 黄光球；

【作者基本信息】 西安建筑科技大学 ， 管理科学与工程， 2004， 硕士

【摘要】 手写字符的识别研究时冷时热，在过去几十年里，人们提出了许多识别方法和识别技术，但由于识别的关键技术没有解决，再加上产品定位等方面的原因，使得已有的识别系统远不能达到实际应用的要求，这其中有理论研究和技术实现等多方面因素。手写体数字识别是字符识别的一个分支，问题虽然简单，但却有较大的实用价值。目前我国在信函通信时广泛使用了邮政编码，用手写体数字识别技术进行信函的自动分拣对减轻邮电职工的手工分拣工作有很大意义。手写数字虽然只有10个种类，但很多情况下，对识别的精度要求非常高，而且，手写数字的变动性非常大，在这种情况下，要想做到高精度的识别就不是那么容易了。 本论文首先阐述计算机字符识别技术的必要性，论述手写数字识别的意义；接着讨论了手写数字识别的预处理技术，包括二值化、行字切分、平滑、去噪声、规范化和细化等。二值化时对整体阈值二值化、局部阈值二值化、动态阈值二值化和利用空间信息进行阈值选取几种常用的阈值选取方法进行讨论，特别对利用空间信息进行阈值选取进行了详细论述；在对通过对基于数学形态学的细化的基础上，讨论序贯同伦形态细化算法和保形的快速形态细化算法；然后依据联机字符识别原理框图，分析了手写数字的结构特点，提出了基于笔划特征的任意手写数字在线识别技术和基于多级分类器任意手写数字在线识别技术，对其中涉及的笔划识别前的噪声处理、笔划间特征量的定义及识别、整字匹配的距离准则进行了详细叙述；继而在对手写数字的分割的基础下对脱机手写数字识别进行了研究，对基于最小距离分类器字符识别、基于树分类器的字符识别、基于自适应共振(ART)网络的字符识别分别进行了详细讨论，并引入置信度分析将多个分类器进行了混合集成；最后简单阐述了手写数字识别的典型应用，对其在大规模数据统计、财务、税务、金融及邮件分拣中的应用进行了探索。 本论文对手写数字识别的原理、方法进行了深入的研究，提出的识别技术精度较高，可以达到实际应用的要求。本论文成果对于信息的自动化、国民经济信息网络的推广具有重要意义，对于手写汉字识别的研究具有很高的参考价值。更多还原

【Abstract】 The recognition research of handwritten character sometimes cold and sometimes hot Over the past dozens of years, people propose a lot of recognition method and recognition technology, but because the key technology of recognition was not solved, in addition, such reasons of the aspect as the products make a reservation, which made the existing recognition system be unable to meet the requirement of practical application. There are factors in many aspects, such as theoretical research and technology, etc. Handwritten digit recognition is one branch of character recognition.Though it is simple, there is greater practical value. At present, zip codes of are extensively used in letter communicating in our country. Automatically sorting letter with handwritten digital recognition technology have very heavy meaning to lightening post worker’s manual sorting. Handwritten numeral has 10 kinds only, but in a lot of situations, recognition precision have very high expectations, in addition, the change of the handwritten numeral is very large. In this case, it is not so easy making sure that high-accuracy recognition.This thesis explains the necessity of the character recognition technology of the computer at first, describe the meaning in which the handwritten numeral discerns; Pretreatment technology of handwritten numeral recognition, including two value, line segmentation, word segmentation smooth, removing noising,standardization and thinning are discussed Two value concretely discusses whole threshold value, some threshold value, dynamic threshold value and utilize space information to carry on threshold, which are several kinds of common method of choosing threshold value, especially utilize space information to carry on threshold value is describe in detail; adopting to the foundation of thinning based on mathematics morphology, Thinning algorithm of serials same and thinning algorithm of protecting shape are discussed; Afterwards, according to principle’s diagram of the on-line character recognition, by analyzing the structure feature of the handwritten numeral, this thesis has proposed the online recognition technology of the free handwritten numeral based on the stroke feature and the online recognition technology of the free handwritten numeral based on the multistage classifying device.Detail narrated noise removing, stroke characteristic definition and discernment, distance criterion of whole word match; then under the foundation of handwritten numeral segmentation, off-line handwritten numeral recognition is researched. Especially minimum distance classifying device, tree classifying device and adaptive resonance (ART) network classifying device is discussed At the same time, believes degree analyses are introduced to integrate a lot of classifying devices; At the end, the typical application of the handwritten numeral recognition was briefly narrated, its application in extensive data statistics, financial affairs, tax, finance and mail sorting have been explored.This thesis deeply researched into the principle and method of handwritten numeral recognition. The recognition technology putted forward is relatively high in precision, can meet the requirement of practical application. This achievement of thesis has significant meaning to the automation of information and popularization of national economic information network, and has very high reference value to the research of handwritten Chinese character recognition.更多还原