人脸识别及其在安卓操作系统中的应用英语论文

HMM-basedfacerecognitionsystemembeddedintheandroidintheresearchanditsapplicationKeywords:HMMfacerecognitionembeddedARM90IntroductionEmbeddedfacerecognitionsystemandcomparedtothetraditionalsystemofidentificationhasastrongadvantage,nospecialacquisitionequipment,lowcost,simpletouse;thesametime,facerecognitiondoesnotinterferewiththeuser,doesnotinfringetheprivacyofusers,areinitiativetoidentifynon-infringement,easyforuserstoaccessthismainlineofembeddedARM9systemdevelopmenttothetheoreticalbasisoftheHMM,showscompleteimagecapture,facedetectionandrecognitionfunctionscorrespondingtothehardwareplatformandsoftwaremodulesdesignedtoandimplementationprocess;andimagepre-processingisoptimizedtodofloating-pointarithmetic,greatlyincreasingthespeedofembeddedsystems.PartofthesystemsoftwarecanbedirectlyappliedwiththeLinuxoperatingsystemforsmartphones,theuseofmobilephonesandbuilt-incamera,apersonalfacialfeaturedatacanbeanalyzedandthenstoredincontrasttotheinitialfaceinformationdatabase,completetheidentificationfunction.1AsystemarchitectureanddesignThesystemusestheSamsungintroducedtheARM920TRISCprocessorkernel-S3C2410A.Itsexcellentprocessingperformanceshouldbeviewedasthefirstchoiceforthedevelopmentofportabledevices.Atthesametimetomeetthesmartphonesonthevideoimageacquisitionneeds,thesystemusesaUSBbus-basedvideoacquisitionmodule,andgreatlyimprovedcomparedtotheserialtransmissionofdataacquisitionrates.Thesysteminvolvesdigitalimageacquisition,processing,storage,transmissionandHMMalgorithmsandothertechniques.ThesystemarchitectureshowninFigure1.2imageacquisitionhardwaredesignInviewofthetraditionalhighcostCCDimagesensor,therelativecomplexityoftheadditionalcircuitryandhighpowerconsumption,thesystemusesOmniVision'sOV7640CMOSchipasthecompany'simagesensor.OV7640isalowvoltage(2.5V),highsensitivityCMOSimagesensor.Real-timeacquisitionandstoragesystemsneedhigh-speeddatatransmission,thesystemhardwarewiththequestionputforwardhigherrequirements.Thesystemdesign,partsacquisitionandtransmissionpartsinbetweenwiththeappropriatebuffer.Practice,andsupportingtheuseofOV7640OV511extendDRAMchipsfromthecacheroleinachievinghigh-speeddigitalvideoimagesviaUSBintotheARMprocessor.OV511isadedicateddigitalcameraUSBinterfaceICchip.3imagecaptureprogramThesystemusesLinuxastheoperatingsystemplatform,operatingsystemmigrationdonotdotoomuchinthisintroduction.Video4Linux(shortV4L)isaLinuxkernelonthevideodevicedriver,itisforthevideoequipmenttoprovideaseriesofapplicationprogramminginterfaceunctions,thevideofequipmentonthemarkettoday,includingthepopularTVcard,videocapturecardsandUSBcameras.Video4LinuxLinuxkernelprovidestheapplicationinterface,programdevelopment,thefirstisbasedontheVideo4LinuxAPIfunctiontodesigntheprogram.Video4LinuximageacquisitionbasedontheprogramflowshowninFigure2.4ImagepreprocessingandfacerecognitionalgorithmandimplementationFacerecognitionprocessfirsttodeterminetheinputfaceimageorfacetheexistenceofthevideo,ifthereisfurthergiventhepositionofeachface,sizeandlocationofeachofthemajororgansoffacialinformation,andbasedonthisinformation,furtherextractionofeachinherentinthefaceofpersonalidentity,andhasbeenthefaceofitslibraryinthefacecomparedtoidentifytheperson'sidentity.Facerecognitionprocesscanbedividedintoimagepre-processing,facedetectionandfacerecognitionofthreeparts.Facedetectionisthepositioningofthematrixtobeidentifiedfromthefaceregioninthefeaturearea,andseparatedfromeachregion.Facerecognitionisbasedontheexistingfacedatabase,theoutputcorrespondingtothetestfaceinthefacedatabaseobjectlabel.Bothaspremiseandpurpose.HMMcanbecompletedasfacedetection,facerecognitioncanbedone,sowewillfacedetectionandrecognitionsimultaneously.4.1HiddenMarkovModel(HMM)basicconceptsHMMisasetofstatisticalpropertiesofthesignalcharacteristicsofthemodel,whichconsistsoftworelatedprocesses:oneistheimplicit,invisibletoafinitestateMarkovchain,whichhasinitialstateprobabilitydistributionfunctionandthestatetransitionprobabilitymatrix,theotherisagroupwithstateoftheprobabilitydensityfunction.4.2HMMmodelforfacerecognitionAHMMcanbedenotedbyλ={A,B,Π},becauseofitslimitedcharactersetinputV={v1,v2,...vm},socalleddiscretehiddenMarkovmodel.Accordingtothetypeofstatetransition,HMMcanbedividedthroughthe(ergodic)andfromlefttoright(left-right).Theformerstatetransitionthatisarbitrary,canbetoownandallotherstate,whichstatetransitionislimitedtoitselfandthenextstate.Faceverticalandhorizontaldirectionfromlefttorightfromtoptobottominallregionswiththesamenaturalorder,canbesimulatedwiththe1D-HMMforface,showninFigure3.4.3facialimagefeatureextractionLeteachindividualfaceimagewidthW,heightH,isdividedintooverlappingblocks.TheheightoftheblockL,overlappingdepthP.Therefore,thefaceimageextractedfromtheblockstoscorethenumberofobservedvectorT,andT=(HL)/(LP)+1.ChoiceofparametersLandPwillaffectthesystem'srecognitionrate,thelargevalueofPincreasedthedepthofoverlappingverticalnumberoffeaturevectors,thesystemrecognitionrate.L,choosemoresubtle,smallerLcannoteffectivelyidentifytheobservationvector;andlargeLtheshearincreasestheprobabilityofint