人脸识别文献翻译(中英文)

附录(原文及译文)翻译原文来自ThomasDavidHeseltineBSc.Hons.TheUniversityofYorkDepartmentofComputerScienceFortheQualificationofPhD.--September2005-《FaceRecognition:Two-DimensionalandThree-DimensionalTechniques》4Two-dimensionalFaceRecognition4.1FeatureLocalizationBeforediscussingthemethodsofcomparingtwofacialimageswenowtakeabrieflookatsomeatthepreliminaryprocessesoffacialfeaturealignment.Thisprocesstypicallyconsistsoftwostages:facedetectionandeyelocalisation.Dependingontheapplication,ifthepositionofthefacewithintheimageisknownbeforehand(foracooperativesubjectinadooraccesssystemforexample)thenthefacedetectionstagecanoftenbeskipped,astheregionofinterestisalreadyknown.Therefore,wediscusseyelocalisationhere,withabriefdiscussionoffacedetectionintheliteraturereview(section3.1.1).Theeyelocalisationmethodisusedtoalignthe2Dfaceimagesofthevarioustestsetsusedthroughoutthissection.However,toensurethatallresultspresentedarerepresentativeofthefacerecognitionaccuracyandnotaproductoftheperformanceoftheeyelocalisationroutine,allimagealignmentsaremanuallycheckedandanyerrorscorrected,priortotestingandevaluation.Wedetectthepositionoftheeyeswithinanimageusingasimpletemplatebasedmethod.Atrainingsetofmanuallypre-alignedimagesoffacesistaken,andeachimagecroppedtoanareaaroundbotheyes.Theaverageimageiscalculatedandusedasatemplate.Figure4-1-Theaverageeyes.Usedasatemplateforeyedetection.Botheyesareincludedinasingletemplate,ratherthanindividuallysearchingforeacheyeinturn,asthecharacteristicsymmetryoftheeyeseithersideofthenose,providesausefulfeaturethathelpsdistinguishbetweentheeyesandotherfalsepositivesthatmaybepickedupinthebackground.Althoughthismethodishighlysusceptibletoscale(i.e.subjectdistancefromthe2camera)andalsointroducestheassumptionthateyesintheimageappearnearhorizontal.Somepreliminaryexperimentationalsorevealsthatitisadvantageoustoincludetheareaofskinjustbeneaththeeyes.Thereasonbeingthatinsomecasestheeyebrowscancloselymatchthetemplate,particularlyifthereareshadowsintheeye-sockets,buttheareaofskinbelowtheeyeshelpstodistinguishtheeyesfromeyebrows(theareajustbelowtheeyebrowscontaineyes,whereastheareabelowtheeyescontainsonlyplainskin).Awindowispassedoverthetestimagesandtheabsolutedifferencetakentothatoftheaverageeyeimageshownabove.Theareaoftheimagewiththelowestdifferenceistakenastheregionofinterestcontainingtheeyes.Applyingthesameprocedureusingasmallertemplateoftheindividualleftandrighteyesthenrefineseacheyeposition.Thisbasictemplate-basedmethodofeyelocalisation,althoughprovidingfairlypreciselocalisations,oftenfailstolocatetheeyescompletely.However,weareabletoimproveperformancebyincludingaweightingscheme.Eyelocalisationisperformedonthesetoftrainingimages,whichisthenseparatedintotwosets:thoseinwhicheyedetectionwassuccessful;andthoseinwhicheyedetectionfailed.Takingthesetofsuccessfullocalisationswecomputetheaveragedistancefromtheeyetemplate(Figure4-2top).Notethattheimageisquitedark,indicatingthatthedetectedeyescorrelatecloselytotheeyetemplate,aswewouldexpect.However,brightpointsdooccurnearthewhitesoftheeye,suggestingthatthisareaisofteninconsistent,varyinggreatlyfromtheaverageeyetemplate.Figure4-2–Distancetotheeyetemplateforsuccessfuldetections(top)indicatingvarianceduetonoiseandfaileddetections(bottom)showingcrediblevarianceduetomiss-detectedfeatures.Inthelowerimage(Figure4-2bottom),wehavetakenthesetoffailedlocalisations(imagesoftheforehead,nose,cheeks,backgroundetc.falselydetectedbythelocalisationroutine)andonceagaincomputedtheaveragedistancefromtheeyetemplate.Thebrightpupilssurroundedbydarkerareasindicatethatafailedmatchisoftenduetothehighcorrelationofthenoseandcheekboneregionsoverwhelmingthepoorlycorrelatedpupils.Wantingtoemphasisethe3differenceofthepupilregionsforthesefailedmatchesandminimisethevarianceofthewhitesoftheeyesforsuccessfulmatches,wedividethelowerimagevaluesbytheupperimagetoproduceaweightsvectorasshowninFigure4-3.Whenappliedtothedifferenceimagebeforesummingatotalerror,thisweightingschemeprovidesamuchimproveddetectionrate.Figure4-3-Eyetemplateweightsusedtogivehigherprioritytothosepixelsthatbestrepresenttheeyes.4.2TheDirectCorrelationApproachWebeginourinvestigationintofacerecognitionwithperhapsthesimplestapproach,knownasthedirectcorrelationmethod(alsoreferredtoastemplatematchingbyBrunelliandPoggio[29])involvingthedirectcomparisonofpixelintensityvaluestakenfromfacialimages.Weusetheterm‘DirectCorrelation’toencompassalltechniquesinwhichfaceimagesarecompareddirectly,withoutanyformofimagespaceanalysis,weightingschemesorfeatureextraction,regardlessofthedistancemetricused.Therefore,wedonotinferthatPearson’scorrelationisappliedasthesimilarityfunction(althoughsuchanapproachwouldobviouslycomeunderourdefinitionofdirectcorrelation).WetypicallyusetheEuclideandistanceasourmetricintheseinvestigations(inverselyrelatedtoPearson’scorrelationandcanbeconsideredasascaleandtranslationsensitiveformofimagecorrelation),asthispersistswiththecontrastmadebetweenimagespaceandsubspaceapproachesinlatersections.Firstly,allfacialimagesmustbealignedsuchthattheeyecentre