欢迎来到天天文库
浏览记录
ID:34017150
大小:2.76 MB
页数:55页
时间:2019-03-03
《基于单演二值模式的微表情识别的研究》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ABSTRACTMicro-expressionisabriefandinvoluntaryfacialexpressionwhichonlycanbedetectedbyafewofpeopleinreallife.Comparedwithnormalfacialexpression,micro-expressionismorelikelytorevealpeople’sfeelingandmotivation.Itusuallyappearswhenpeopletrytoconcealorrepresstheirrealemotions,somicro-exp
2、ressionisaneffectiveclueforlieindication.Asforitsshortdurationandlowintensity,theresearchofautomaticmicro-expressiondetectionandrecognitionisachallengingtask.Thispaperpresentsamethodformicro-expressionfeatureextractionbaseonmonogeniclocalbinarypattern,andappliesittodetectmicro-express
3、ioninstaticimagesanddynamicimagesequences.Themainworkisasfollows:(1)Faciallandmarklocation.Asthecomputationinparameterizedappearancemodels(e.g.,ActiveAppearanceModel)isexpensiveandcomplexity,thispaperpresentsthesuperviseddescentmethod(SDM).SDMextractfeaturesbylearningthedescentdirecti
4、onsofthesequencesinasupervisedmanner,whichreducesthecalculation..(2)Thispaperpresentsthemonogenicbinarypattern(MBP)formicro-expressionfeatureextraction.Thismethodusesfewerconvolutionstoextractmorecompactfeaturevectors,whichreducestimeandspacecomplexity.Dynamicfeaturecanrepresentmoreco
5、mprehensiveinformationofmicro-expression,soMBPisexpandedtothreeorthogonalplanesMBPtoclassifybetter.(3)Asprobabilitystatisticsmodels,HiddenMarkovModelshasbeenappliedinspeechrecognitionandfacialexpressionrecognitionsuccessfully.AmethodbasedontheHiddenMarkovModelsispresentedherewhichgain
6、saHMMforeachmicro-expression.TheexperimentssimulatebyMATLAB.Therearetwomicro-expressionimagedatabasesusedintheexperiments,CASMEandSMIC.ExperimentalresultsshowthattheproposedmethodhasbetterperformancethanGaborandreducethecomplexityintimeandspace.KEYWORDS:micro-expression,superviseddesc
7、entmethod,monogenicsignal,MBP,HiddenMarkovModelsII目录第一章绪论......................................................................................................................-1-1.1课题研究背景与意义...............................................................................................
8、..-1-
此文档下载收益归作者所有