资源描述：
《Face Recognition Based on Fitting a 3D Morphable Model》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、IEEETRANSACTIONSONPATTERNANALYSISANDMACHINEINTELLIGENCE,VOL.25,NO.9,SEPTEMBER20031FaceRecognitionBasedonFittinga3DMorphableModelVolkerBlanzandThomasVetter,Member,IEEEAbstractÐThispaperpresentsamethodforfacerecognitionacrossvariationsinpose,rangingfromfrontaltoprofileviews,andac
2、rossawiderangeofilluminations,includingcastshadowsandspecularreflections.Toaccountforthesevariations,thealgorithmsimulatestheprocessofimageformationin3Dspace,usingcomputergraphics,anditestimates3Dshapeandtextureoffacesfromsingleimages.Theestimateisachievedbyfittingastatistical,
3、morphablemodelof3Dfacestoimages.Themodelislearnedfromasetoftextured3Dscansofheads.Wedescribetheconstructionofthemorphablemodel,analgorithmtofitthemodeltoimages,andaframeworkforfaceidentification.Inthisframework,facesarerepresentedbymodelparametersfor3Dshapeandtexture.Wepresentr
4、esultsobtainedwith4,488imagesfromthepubliclyavailableCMU-PIEdatabaseand1,940imagesfromtheFERETdatabase.IndexTermsÐFacerecognition,shapeestimation,deformablemodel,3Dfaces,poseinvariance,illuminationinvariance.æ1INTRODUCTIONNfacerecognitionfromimages,thegray-levelorcolorTwo-dimen
5、sionalfacemodelsrepresentgrayvaluesIvaluesprovidedtotherecognitionsystemdependnotandtheirimagelocationsindependently[3],[4],[18],[23],onlyontheidentityoftheperson,butalsoonparameters[13],[22].Thesemodels,however,donotdistinguishsuchasheadposeandillumination.Variationsinposeandb
6、etweenrotationangleandshape,andonlysomeofthemillumination,whichmayproducechangeslargerthantheseparateilluminationfromtexture[18].Sincelargerota-differencesbetweendifferentpeople'simages,arethemaintionscannotbegeneratedeasilybythe2Dwarpingusedchallengeforfacerecognition[39].Theg
7、oalofrecognitioninthesealgorithmsduetoocclusions,multipleview-basedalgorithmsistoseparatethecharacteristicsofaface,which2Dmodelshavetobecombined[36],[11].Anotheraredeterminedbytheintrinsicshapeandcolor(texture)ofapproachthatseparatestheimagelocationsoffacialthefacialsurface,fro
8、mtherandomconditionsofimagefeaturesfromtheirappearance
