基于皮肤模板和改进HMM的自动人脸识别系统

《基于皮肤模板和改进HMM的自动人脸识别系统》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、维普资讯http://www.cqvip.com第25卷第1期深圳大学学报理工版Vo1．25No．12008年1月JOURNALOFSHENZHENUNIVERSITYSCIENCEANDENGINEERINGJan．2008文章编号：1000—2618(2008)01—0071—05【电子光学与信息工程】A1uttomaticflacereco,tg~nitionbasedoonskK,inmaski"ngandimprovedHMMSHENLin—linandMINGZhongCollegeofSoftwareShenzhenUniversityShenzhen

2、518060P．R．ChinaAbstract：AnewhiddenMarkovmodule(HMM)basedfacerecognitionsystemispresentedinthispaper．FaceimageswereextractedautomaticallyfromlivevideoscapturedbyaCreativeWebCam，andfaceswererecognizedinrealtimeusinganimprovedHMMfacerecognitionalgorithm．Afastfacedetectionalgorithmusingboo

3、stedHaar—likefeatureswasappliedinitiallytodetectfacesregionsfromthevideostream．Thedetectedfaceregionwasfurtherre—finedbyaskincolormaskingmoduletoachievemoreaccuratefaceposition．ToimprovetheaccuracyofHMMbasedfacerecognitionalgorithm，discretewavelettransform，insteadofdiscretecosinetransf

4、orm，wasusedtoextractob．servationsequencesforHMM．Experimentswereconductedusingtwofacedatabases：theORLdatabaseandtheNottinghamcolorfaceimagedatabase．Theresultsonbothdatabasesshowthattheproposedmethodcanimprovetheaccuracybymorethansixpercents．Furtherimprovementhasbeenobservedwhenaskinmask

5、ingmoduleisusedtorefinethedetectedfaceregion．Keywords：facerecognition；facedetection；colorimageprocessing；biometrics；hiddenMarkovmodelCLCNumber：TP391；TP751Documentcode：AAutomatedfacerecognitionsystemshaveawiderangeofim—thesystem．Itsaccuracyisimprovedusingafastskincolorbasedportantapplic

6、ationssuchasvideosurveillance，crimepreventionmaskasapostprocessingproceduretolocatethefaceareaaccu—anddetection．Acompletefacialimagerecognitionsystemshouldrately．Duetotheaccuracyofthedetectedfaces，thedetectedbeabletodetectfacesinagivenimageandrecognizethematthefacescanbeeasilynormalize

7、d．sametime．Oncethefacepositionisdetermined，facialareawillMajorapproachestofacerecognitionincludegraphmate—beextractedfroIntheimage，normalizedinbothscaleandorien—hing，neuralnetworksandhiddenMarkovmodels(HMM)．Antation，andpassedontothefacerecognizer．exampleofgraphmatchingisthedynamiclin