基于SVM和BP神经网络的手写数字识别

I摘要科技发展日新月异，智能识别推陈出新。如今是信息化时期，数字识别在很多智能领域上运用广泛，拥有普遍的使用远景，因此探索这项技术有其重要的实际意义。由于手写数字在写法上千差万别，且数字间字形差别相对较小，使得识别系统的开发具有很大的挑战性。当前手写数字识别采用的技术有Bayes判别法、决策树法、神经网络和支持向量机（SupportVectorMachines，SVM）等。诞生于20世纪90年代的SVM技术是机器学习研究的热点，因其良好的泛化性能成为了数字识别领域的热门方法。本开发系统借助MATLAB平台实现完成SVM的手写数字识别功能，同时与BP神经网络的识别作对比，并利用了MNIST数据库作扩展与分析。对识别的结果进行探究，得出使识别精准度出现误差的主要因素有手写体数字的规范程度、笔画字迹粗细和清晰，以及训练样本的数量等。关键词手写数字识别；神经网络；SVMAbstractIITechnologicaldevelopmentchangesrapidly,andintelligentrecognitioninnovatesconstantly.Intheinformationera,numeralrecognitionhasbroadapplicationprospectsinmanypatternareaswithacommonvision,soitisofgreatpracticalsignificancetoexplorethistechnology.Asaresultofhandwrittennumeralvarywidelyinthewordinganddigitalshapedifferenceisrelativelysmall,makingrecognitionsystemdevelopmentisagreatchallenge.Atpresent,Bayesdiscriminantanalysis,decisiontreemethod,neuralnetworkandsupportvectormachine(referredtoasSVM)arethemainmethodsofrecognitionofhandwriting.SVMtechnology,whichwasbornin1990s,isahottopicinmachinelearningresearch.Becauseofitsgoodgeneralizationperformance,ithasbecomeapopularmethodinthefieldofnumeralrecognition.ThesystemusestheMATLABtorealizethehandwrittennumeralrecognitionbasedonSVM,atthesametimewiththerecognitionofBPneuralnetworkforcomparison,andusingtheMNISTdatabasetoextendandanalysis.Toresearchtheresultsofrecognition,itcomestoconclusionthatthemainfactorsaffectingtheaccuracyoftherecognitionincludethespecificationofthehandwrittennumeral,thethicknessandtheclarityofthestrokes,andthenumberofthetrainingsamples.Keywordshandwrittennumeralrecognition;neuralnetwork;SVMIII目录摘要............................................................................................................................................................IAbstract............................................................................................................................................................I第1章绪论.....................................................................................................................................................11.1课题研究的背景和意义...............................................................................................................11.2国内外研究现状及分析...............................................................................................................11.3课题研究的主要内容....................................................................................................................11.4本章小结..........................................................................................................................................2第2章手写数字识别综述..........................................................................................错误!未定义书签。2.1预处理技术.......................................................................................................错误!未定义书签。2.1.1图像二值化..........................................................................................错误!未定义书签。2.1.2图像去噪锐化......................................................................................错误!未定义书签。2.1.3图像分割细化......................................................................................错误!未定义书签。2.1.4图像归一化..........................................................................................错误!未定义书签。2.2特征提取技术.................................................................................................................................32.3手写数字识别方法........................................................................................................................42.3.1决策树法..............................................................................................................................42.3.2贝叶斯判别法......................................................................................................................52.3.3神经网络..............................................................................................................................72.3.4支持向量机.........................................................................................................................92.4本章小结..........................................................................................................................................9第3章支持向量机.......................................................................................................................................93.1SVM概述............................................................................................................................................93.1.1VC维....................................................................................................................................103.1.2结构风险最小原理..........................................................................................................103.2SVM的原理......................................................................................................................................113.2.1二分类支持向量机...................................................................................................