一种基于局部特征分析的人脸识别方法

11111,211(ICT-YCNC100080)2(150001)GaborGabor,,Gabor,1,[1][2][3](Eigenface)[4],[5]Gabor,2.2,Gabor,Gabor[6],33.1AABAHHLR1)LRH2)LR3)LR4)LR5)LR1GaborGabor,3LRBEΩ==∑Ω∈8)y(x,,,0),(|),(),(),(θjijiEyxByxRng(1)P∑∈=AyxyxpENS),(),(1(2)EA21RrR≤≤1R2RN)arg(pRngpLeftIrisSMaxPleft∈=(3))arg(pRngpRightIrisSMaxPRight∈=(4)3.23.30.420.510.881.590.570.700.270.350.030.050.040.090.050.060.020.05143.3.123))2/(exp()(:)()()(*)()(22sxxgjigifigifihj−=−==∑(8))(2)1()1()(2/))1()1(()(ihihihihihihih−−++=′′−−+=′(9)}0)(&&0)1()1(|{=′′=+′+′=ihihihiS(10)}),({minargSiifmi∈∀=(11)232-323[6][7]26105113545[2]5(4)(5)44.1Gabor1964GaborGaborGabor[8]GaborGaborDaugman80[9]Gabor()−−−=2expexp2exp22222σσϕxkixkkxjjjjrrr(12)5,v=0,1,2,3,48u=0,…,7()()uvuvjxjykkkkjkφφcossin==rπ222+−=vvk8πφuu=→xL),(yxx=→Gabor()()()'2''xdxxxLxJjjrrrrr−=∫ϕ(13)Gabor4058()401exp≤≤=jiaJjjjφ'JJ,6()∑∑∑=jjjjjjjaaaaaJJS2'2'',(14)aS()∑∑∑−−=jjjjjjjjjjaakdaaJJS2'2'''cos,rrφφφ,(15)dr'JJΦSJ'Jdr8dr[5]P,'PGaborpJGabor'pJP,64.2(6)(FaceBunchGraph,FBG)[5]FBGnJGaborGabor7FBGeX→∆60GaborkiJK601K=K,i261K=iFBGeX∆4.310PGabor0J2GaborkJ,601K=K,0JkJ()kkJJd,0r3kkdPPr+=0'Gabor'kJ414'kJkJ()kkkaJJS,''kP884.4(1)()()∑∑==∆∆−∆−=EeFBGeFBGeIeFBGKIkLkiaXXXEJJaxSmLFBGISi1221)(),(1,λφ(16)LEλ262.8[5],85.Gabor{kiJJJL,,21},K26GaborNiJGabor'NiJNiJ'NiJ()()∑−=iiieyxyxd2,rr(17)()⋅=−yxyxyxdabcrrrrrr1cos,(18)()∑−=iicbdyxyxdrr,(19)3007Eigenface(48)9LocalfeaturemethodEigen+NNDeDabcDcbd92.3%92.7%93.3%86.7%21OlivettiORL40,8ORLLocalfeaturemethodEigen+KNNDeDabcDcbd93.75%93.5%94.5%91.5%2EigenfaceEigenface6.[5]4%3003003,ORL4002310EigenFaceEigenFace[1]R.Chellappa,C.L.Wilson,andS.Sirohey.Hunamandmachinerecognitonoffaces:Aservey.Proc.IEEE,83:705-741,May1995[2]W.Zhao,R.Chellappa,A.Rosenfeld,P.J.PhillipsFaceRecognition:ALiteratureSurvey.url=citeseer.nj.nec.com/374297.html[3]R.BrunelliandT.Poggio.Facerecogniton:Featuresversustemplates.IEEETransactionsonPatternAnalysisandMachineIntelligence,15:1042-1052,1993.[4]TurkM,PentlandA.Eigen-facesforRecognitionJournalofcognitiveneuroscience,3(1),pp71-86,1991[5]LaurenzWiskott,jean-MarcFellous,NorbertKruger,andChristophvonderMalsburg,FaceRecognitionbyElasticGraphMatching,IEEETransationsonPatternAnalysisandMachineIntelligence,Vol.19,no,7,July1997.[6]liu,ming-bao.Studyonhumanfacedetectionandtraching[PH.D.Thesis].Harbin:HarbinInstitudeofTechnology,1997(inChinese).[7]ShanShiguang,Gaowen,ChenXilin.FacialFeatureExtractionBasedonFacialTextureDistributionandDeformableTemplate.JournalofSoftware,Vol.12No.4April2001,571-577.[8]R.N.Bracewell,TheFourierTransformandItsApplication,NewYork,McGraw-Hill,1978[9]Daugman,J.G.(1998).Completediscrete2-DGabortransformbyneuralnetworkforimageanalysisandcompression.IEEETrans.OnAcoustics,SpeechandSignalProcessing,36(7):1169-1179AbstractElasticBunchGraphMatchinghasbeenprovedeffectiveforfacerecognition.Buttherecognitionprocedureneedslargecomputation.Herewepresentanautomaticfacerecognitionmethodbasedonlocalfeatureanalysis.Thelocalfeaturesarefirstlylocatedbythefacestructureknowledgeandgrayleveldistributioninformation,ratherthansearchingonthewholeimageasitdoesinElasticBunchGraphMatching.Theroughpositionsofthefeaturesarelocated.ThenweuseadatastructurenamedFaceBunchGraphtoadjustthefeaturepositions.Aftertheaccuratepositionsofthefeaturesarelocated,theGaborjetsonthesepointsarecalculated.Thefaceis11representedbyGaborjetsofthefeaturesandtheirspatialdistances.Severaldistancemetricsaretestedandtheresultsaregiven.ComparedtothetraditionalElasticBunchGraphmethod,thelocalfeaturesarefirstlylocatedratherthansearchingonthewholeimage,thewholecomputationisgreatlydecreased.KeyWordsFacerecognition,Featurelocation,Gaborjets,Facebunchgraph(69789301)863(863-306-ZT03-01-2)[e-mail]82649018