基于wmcbp-wwef特征融合的随机森林人脸表情识别-random forest peoples facial expression recognition based on wmc bp - wwef feature fusion.docxVIP

FacialExpressionRecognitionBasedonWMCBP-WWEF FeatureFusionUsingRandomForestABSTRACTFacialexpressionrecognitionisahotresearchtopicofcomputervisionand patternrecognition,causeswidespreadconcernof an increasing number of domesticandforeignresearchers.Thegoaloffacialexpressionrecognitionisto makeartificialintelligencecanautomaticallyidentifypersonsexpression,andthen analyzetheirinneremotions.It canenhancethefriendlinessandintelligenceof human-computerinteraction,soitwillhasabroadapplicationprospectsand potentialmarketvalue.Theexpressionrecognitionsystemmainlyconsistsofthefollowingparts:face detection,imagepreprocessing,expressionfeatureextractionandexpression classification.Thisdissertationresearchesonsomekeyissuesofthestaticimage expressionrecognition,improvessome algorithms,and verifies by experiments. Themainworkandinnovationsareasfollows:Expressionimagepreprocessing.Duetoavarietyofimagesize,angle,and theinfluenceoftheilluminationcondition, theimageispreprocessedbyusing mean-variancenormalizationandgeometricnormalization.Successfulpretreatment haslaidagoodfoundationforimprovingtheexpressionrecognitionrate.Featureextraction.Todealwiththedeficiencyoftheexistingmulti-scale centralizedbinarypatterns,afeatureextractionmethodcalledwaveletmulti-scale centralizedbinarypattern(WMCBP)isproposedbasedonwavelettransformation. Thismethodcannotonlygetmoreaccuratemulti-scaleinformation,butalso greatlyreducethecomputationalcomplexity.Weightedwaveletenergyfeature (WWEF)isalsointroducedtogetthedetailsofthesub-pictureinformationinorder tofurtherenhancethefacialexpressionrecognitionaccuracy.Thesetwotypesof featuresarecomplementarytosomeextent,andthefusionofthemcanenhancethe performanceofWMCBPinfacialexpressionrecognitionwithoutincreasingthe computationobviously.Expression classification.Asafamousensemblelearningclassifier, randomforestalgorithm(RF)isoneoftheclassicclassificationmethodsinpattern recognition.RFcaneffectivelydealwithlargedatasets,givethebettervariable importanceestimationin