搜索
摘要工业大学学位论文I摘要人脸表情识别技术是人工智能领域新兴的一个研究方向，它在智能化的人际交互中具有广阔的应用前景。同时，该技术还被广泛应用在交通、医疗和公共安全等方面。近年来，人脸表情识别技术受到了更多学者的关注，成为了人工智能领域的又一研究热点。因此，对人脸表情识别技术的研究具有重要的理论意义和实际的应用价值。人脸表情识别技术主要包括图像预处理、特征提取和分类识别。在人脸表情图像预处理中，针对人眼定位的问题，给出了一种基于Sobel边缘提取的人眼定位新方法。主要包括眼部粗定位、眉眼分离和精确定位三个步骤。眼部粗定位是在对图像进行中值滤波和归一化预处理后，根据先验知识来完成。眉眼分离是利用灰度积分投影曲线来确定眉眼分离线，然后进行分离。眼睛的精确定位是通过对眼睛区域应用Sobel算子提取边缘后，根据二值图像的边界来确定精确的位置。该方法与Hough变换圆检测法和传统模板匹配法相比较，在计算时间上具有明显优势，适合应用于实时人脸表情识别系统中。针对特征提取，给出一种应用Gabor小波变换和非负矩阵分解相结合的特征提取方法。结合Gabor小波变换的特性，设计了相应的Gabor滤波器组，利用它对面部信息区域进行滤波处理，获取不同的子图信息。接着，对各个滤波器滤波产生的子图分别进行非负矩阵分解，实现数据的降维及特征的选择。针对分类识别，设计了基于最近邻思想和概率统计原理的两级分类模式。从各个滤波器获得的子图信息为输入信息，将其输入作为第一级分类器的最近邻分类器，对第一级分类器的输出结果进行概率统计，以实现第二层分类，第二层分类器的输出结果作为最终的识别结果。两级分类模式使每幅图像实现了多次判别，降低了错误分类的可能性，提高了算法的鲁棒性。在MATLAB编程环境下，采用日本女性人脸表情数据库（JAFFE）进行了实验测试。实验结果表明了本文所给方法的有效性。除此以外，还将所给出的方法应用于小样本下的人脸识别中，并在Yale人脸库中进行了实验，结果显示该方法也适用于小样本的人脸识别技术中。关键词：表情识别；人脸识别；非负矩阵分解；Gabor小波变换Abstract工业大学学位论文IIAbstractFacialexpressionrecognitionisanewresearchdirection,andithasbroadapplicationprospectsintheintelligenthuman-computerinteraction.Atthesametime,facialexpressionrecognitioniswidelyappliedinthefieldsofthetraffic,medicaltreatmentandpublicsafety.Inrecentyears,facialexpressionrecognitionhasbeenpaidcloseattentionbymoreandmorescholars.Andthistechnologyhasbecomeahotresearchintheareaofartificialintelligence.Therefore,researchonfacialexpressionrecognitiontechnologyhasimportanttheoreticalsignificanceandpracticalapplicationvalue.Facialexpressionrecognitiontechnologyincludesimagepreprocessing,featureextractionandclassification.Inimagepreprocessing,aneweyelocationmethodbasedonSobeledgeextractionwasproposed.Thenewmethodincludesthreestepsforcoarsepositioningofeyes,segmentationofeyebrowsandeyes,andprecisepositionofeyes.Aftermedianfilteringandnormalizationforimage,thecoarseareaofeyewaslocatedaccordingtotheprioriknowledge.Thesegmentationofeyebrowsandeyeswasimplementedaccordingtothesegmentationlineofeyebrowsandeyes,andthelinewasdeterminedbythegrayintegralprojectioncurve.Then,thepreciselocationoftheeyeswasrealizedaccordingtotheboundaryofbinaryimageobtainedbyextractingeyeedgewithSobeloperator.ComparedwiththetraditionaltemplatematchingandcircleexaminationofHoughtransform,themethodproposedhasobviousadvantagesincomputingtimeandissuitableforreal-timesystem.Infeatureextraction,thefeatureextractionmethodbasedontheGaborwavelettransformandNon-negativeMatrixFactorizationwasproposed.ConsideringthecharacteristicsofGaborwavelettransform，correspondingGaborfiltersweredesigned.ThefacialinformationareawasfilteredbyGaborfilters,anddifferentsub-pictureinformationwasobtained.Then,sub-picturesgeneratedfromeachfilterweredecomposedbynon-negativematrixfactorization,implementingdatadimensionalityreductionandfeatureselectionprocess.Inclassification,thetwo-layerclassificationmodelbasedonnearestneighborhoodclassifierandtheprobabilitystatisticswasdesigned.Asinputinformation,sub-pictureinformationachievedfromeachfilterwasinputthenearestneighborhoodclassifier,whichwasregardasthefirstlayerclassifier.Theoutputresultsfromthefirstlayerclassificationwerecalculatedprobabilitytoimplementthesecondlayerclassifier.Theoutputresultsofthesecondlayerwerethefinalresults.Underthedesignofthetwo-layerclassificationmodel,eachimagewasdistinguishedtwotimes,reducingthepossibilityofmisclassificationandimprovingtherobustnessofthealgorithm.UndertheMATLABprogrammingenvironment,theJAFFEfacialexpressionsdatabasewasappliedinexperiments.Experimentalresultsshowedthismethodiseffective.Inaddition,themethodhadalsobeenappliedinfacerecognitionwithsmallsamples.RelatedexperimentsAbstract工业大学学位论文IIIweredoneintheYalefacedatabase,andtheresultsshowedthemethodproposedisalsosuitableforfacerecognition.Keywords:facialexpressionrecognition;facerecognition;non-negativematrixfactorization;Gaborwavelettransform目录工业大学学位论文IV目录摘要.........................................................................................................................................IAbstract.....................................................................................................................................II1绪论........................................................................................................................................11.1论文的研究背景与选题意义......................................................................................11.2国内外研究现状..........................................................................................................21.3论文的主要内容及结构安排......................................................................................42人脸表情识别概述................................................................................................................62.1人脸表情识别的一般过程..........................................................................................62.2人脸表情特征提取常用方法......................................................................................62.3人脸表情分类识别常用方法......................................................................................82.4人脸数据库介绍.................................