人脸识别论文文献翻译中英文.docx

人脸识别论文文献翻译中英文.docx

《人脸识别论文文献翻译中英文.docx》由会员分享，可在线阅读，更多相关《人脸识别论文文献翻译中英文.docx（20页珍藏版）》请在冰豆网上搜索。

人脸识别论文文献翻译中英文

附录（原文及译文）

翻译原文来自

ThomasDavidHeseltineBSc.Hons.TheUniversityofYork

DepartmentofComputerScienee

FortheQualificationofPhD.--September2005-

《FaceRecognition:

Two-DimensionalandThree-DimensionalTechnique》

4Two-dimensionalFaceRecognition

4.1FeatureLocalization

Beforediscussingthemethodsofcomparingtwofacialimageswenowtakeabrieflookatsomeatthepreliminaryprocessesoffacialfeaturealignment.Thisprocesstypicallyconsistsoftwostages:

facedetectionandeyelocalisation.Dependingontheapplication,ifthepositionofthefacewithintheimageisknownbeforehand（foracooperativesubjectinadooraccesssystemforexample）thenthefacedetectionstagecanoftenbeskipped,astheregionofinterestisalreadyknown.Therefore,wediscusseyelocalisationhere,withabriefdiscussionoffacedetectionintheliteraturereview（section3.1.1）.

Theeyelocalisationmethodisusedtoalignthe2Dfaceimagesofthevarioustestsetsusedthroughoutthissection.However,toensurethatallresultspresentedarerepresentativeofthefacerecognitionaccuracyandnotaproductoftheperformaneeoftheeyelocalisationroutine,allimagealignmentsaremanuallycheckedandanyerrorscorrected,priortotestingandevaluation.

Wedetectthepositionoftheeyeswithinanimageusingasimpletemplatebasedmethod.Atrainingsetofmanuallypre-alignedimagesoffacesistaken,andeachimagecroppedtoanareaaroundbotheyes.Theaverageimageiscalculatedandusedasatemplate.

Figure4-1-Theaverageeyes.Usedasatemplateforeyedetection.

Botheyesareincludedinasingletemplate,ratherthanindividuallysearchingforeacheyeinturn,asthecharacteristicsymmetryoftheeyeseithersideofthenose,providesausefulfeaturethathelpsdistinguishbetweentheeyesandotherfalsepositivesthatmaybepickedupinthebackground.Althoughthismethodishighlysusceptibletoscale（i.e.subjectdistancefromthecamera）andalsointroducestheassumptionthateyesintheimageappearnearhorizontai.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.Wantingtoemphasisethediffereneeofthepupilregionsforthesefailedmatchesandminimisethevarianeeofthewhitesoftheeyesforsuccessfulmatches,wedividethelowerimagevaluesbytheupperimagetoproduceaweightsvectorasshowninFigure4-3.Whenappliedtothediffereneeimagebeforesummingatotalerror,thisweightingschemeprovidesamuchimproveddetectionrate.

Figure4-3-Eyetemplateweightsusedtogivehigherprioritytothosepixelsthatbestrepresenttheeyes.

4.2TheDirectCorrelationApproach

Webeginourinvestigationintofacerecognitionwithperhapsthesimplestapproach,knownasthedirectcorrelationmethod（alsoreferredtoastemplatematchingbyBrunelliandPoggio[29]）involvingthedirectcomparisonofpixelintensityvaluestakenfromfacialimages.Weusetheterm‘DirectCorrelation'toencompassalltechniquesinwhichfaceimagesarecompared

directly,withoutanyformofimagespaceanalysis,weightingschemesorfeatureextraction,regardlessofthedstaneemetricused.Therefore,wedonotinferthatPearson'scorrelat

appliedasthesimilarityfunction（althoughsuchanapproachwouldobviouslycomeunderourdefinitionofdirectcorrelation）.WetypicallyusetheEuclideandistanceasourmetricintheseinvestigations（inverselyrelatedtoPearson'scorrelationandcanbeconsideredasascale

translationsensitiveformofimagecorrelation）,asthispersistswiththecontrastmadebetweenimagespaceandsubspaceapproachesinlatersections.

Firstly,allfacialimagesmustbealignedsuchthattheeyecentresarelocatedattwospecifiedpixelcoordinatesandtheimagecroppedtoremoveanybackgroundinformation.Theseimagesarestoredasgreyscalebitmapsof65by82pixelsandpriortorecognitionconvertedintoavectorof5330elements（eachelementcontainingthecorrespondingpixelintensityvalue）.Eachcorrespondingvectorcanbethoughtofasdescribingapointwithina5330dimensionalimagespace.Thissimpleprinciplecaneasilybeextendedtomuchlargerimages:

a256by256pixelimageoccupiesasinglepointin65,536-dimensionalimagespaceandagain,similarimagesoccupyclosepointswithinthatspace.Likewise,similarfacesarelocatedclosetogetherwithintheimagespace,whiledissimilarfacesarespacedfarapart.CalculatingtheEuclideandistaneed,betweentwofacialimagevectors（oftenreferredtoasthequeryimageq,andgalleryimageg）,wegetanindicationofsimilarity.Athresholdisthenappliedtomakethefinalverificationdecision.

dqg（dthreshold?

accept）（dthreshold?

reject）.Equ.4-1

4.2.1VerificationTests

Theprimaryconcerninanyfacerecognitionsystemisitsabilitytocorrectlyverifya

claimedidentityordetermineaperson'smostlikelyidentityfromasetofpotentialmatchesinadatabase.Inordertoassessagivensystem'sabilitytoperformthesetasks,avarietyof

evaluationmethodologieshavearisen.Someoftheseanalysismethodssimulateaspecificmodeofoperation（i.e.securesiteaccessorsurveillanee）,whileothersprovideamoremathematicaldescriptionofdatadistributioninsomeclassificationspace.Inaddition,theresultsgeneratedfromeachanalysismethodmaybepresentedinavarietyofformats.Throughouttheexperimentationsinthisthesis,weprimarilyusetheverificationtestasourmethodofanalysisandcomparison,althoughwealsouseFisherLinearDiscriminanttoanalyseindividualsubspacecomponentsinsection7andtheidentificationtestforthefinalevaluationsdescribedinsection8.Theverificationtestmeasuresasystem'sabilitytocorrectlyacceptorrejecttheproposedidentityofanindividual.Atafunctionallevel,thisreducestotwoimagesbeingpresentedforcomparison,forwhichthesystemmustreturneitheranacceptanee（thetwoimagesareofthesameperson）orrejection（thetwoimagesareofdifferentpeople）.Thetestisdesignedtosimulatetheapplicationareaofsecuresiteaccess.Inthisscenario,asubjectwillpresentsomeformofidentificationatapointofentry,perhapsasaswipecard,proximitychiporPINnumber.Thisnumberisthenusedtoretrieveastoredimagefromadatabaseofknownsubjects（oftenreferredtoasthetargetorgalleryimage）andcomparedwithaliveimagecapturedatthepointofentry（thequeryimage）.Accessisthengranteddependingontheacceptance/rejectiondecision.

Theresultsofthetestarecalculatedaccordingtohowmanytimestheaccept/rejectdecisionismadecorrectly.Inordertoexecutethistestwemustfirstdefineourtestsetoffaceimages.Althoughthenumberofimagesinthetestsetdoesnotaffecttheresultsproduced（astheerrorratesarespecifiedaspercentagesofimagecomparisons）,itisimportanttoensurethatthetestsetissufficientlylargesuchthatstatisticalanomaliesbecomeinsignificant（forexample,acoupleofbadlyalignedimagesmatchingwell）.Also,thetypeofimages（highvariationinlighting,partialocclusionsetc.）willsignificantlyaltertheresultsofthetest.Therefore,inordertocomparemultiplefacerecognitionsystems,theymustbeappliedtothesametestset.

However,itshouldalsobenotedthatiftheresultsaretoberepresentativeofsystemperformanceinarealworldsituation,thenthetestd