Face_Detection_人脸识别_综述ppt课件.ppt

《Face_Detection_人脸识别_综述ppt课件.ppt》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、FacedetectionandrecognitionComputerVisionBuilding3DmodelReconstructing3DshapeAnalyzingscenesRecognitionWhyisrecognitionsohardOcclusionDifferentPosesVariabilityintrinsicwithinaclassRecognitionObjectDetectionInstanceRecognitionCategoryRecognitionFaceDetection&FaceRecognitionFaceDetectionIdentifya

2、ndlocatehumanfacesinanimageregardlessofPositionScaleRotationOrientationPoseIlluminationDifficultiesPose&RotationStructuralcomponentscomponentsFacialexpressionOcclusionImagingconditionsInOneFaceImage19*19facepattern256361=22888possiblecombinationofgrayvaluesExtremelyhighdimensionalspaceFaceDetec

3、tionKnowledgeBasedFeatureBasedTemplateBasedAppearanceBased……KnowledgeBasedTheserule-basedmethodsencodehumanknowledgeofwhatconstitutesatypicalface.Usually,therulescapturetherelationshipsbetweenfacialfeatures.Thesemethodsaredesignedmainlyforfacelocalization.KnowledgeBasedRules->Eyes,Nose,Mouth…Ho

4、wtoDefineRules?DetailorGeneral?DifferentPose?KnowledgeBased(1)3LevelswithdifferResolutionLowestMediumHighestGeneralDetailFaceCandidateFacialfeaturesKnowledgeBased(2)UseIntensityvalueI(x,y)VI(y)=HI(x)=Multi-Face;ComplexBackgroundKnowledgeBasedSummarySimplerulestodescribefacefeaturesRule-base

5、d;FindfacecandidateWorkwellonunclutteredbackgroundRules:General?orDetail?DifferentPoseFeatureBasedFacialFeatureApplya2ndderivativeGaussianfiltertosearchforinterestpointsTextureDistincttexture:skin,hair,othersSkinColorDefinearegionofskintonepixelsusingCrCbvaluesFeatureBasedSummaryFeaturesareinva

6、rianttoposeandorientationchangeDifficulttolocatefacialfeaturesduetoseveralcorruptionDifficulttodetectfeaturesincomplexbackgroundTemplateBasedStandardfacepatternismanuallypredefinedorparameterizedbyafunction.Givenaninputimage,thecorrelationvalueswiththestandardpatternsarecomputedforthefacecontou

7、r,eyes,nose,andmouthindependently.TheexistenceofafaceisdeterminedbasedonthecorrelationvaluesTemplateBasedIndividualbrightnessofdifferentpartsoffaces(suchaseyes,cheeks,andforehead)Relativebrightnessofthesepartsremainlargelyunchange