基于分块小波包变换和2d2pca的人脸识别算法-face recognition algorithm based on block wavelet packet transform and 2d2pca.docxVIP

AbstractFacerecognitiontechnologyhasbecomeoneofthemosttopicsinbiometricidentificationtechnology,andit’sintuitive,easytoaccept,meanwhile,itcanalsomakehiddenoperations.Becauseofthecomplexityofhumanfaceanditsenvironment,aswellasthediversityoffacialexpressions,facerecognitiontechnologyisstillachallengingresearchtopic.Recently,avarietyoffacerecognitionalgorithmshavebeenproposed,andtheyarewidelyusedinthefacerecognitiontechnology.Thekeypointiswhetherthefeaturesoffaceimagecanbeeffectivelyextracted.Fordifferentfacerecognitionsystemscorrespondingwithdifferentfaceimagefeatures,therecognitionaccuracy,computationalcomplexityandrobustnessaredifferent.Thispapersystematicallysummarizestheresearchachievementsofpredecessorsandtheadvantagesanddisadvantagesofvariousalgorithms.Andonthebasisofit,anewfacerecognitionalgorithmbasedonblockwaveletpackettransformisproposed.Inthispaper,themaintasksareasfollows:Theideaofimagesegmentationisappliedtofacerecognition.Thefaceimageafterpreprocessingisdividedintoanappropriatenumberofsub-blockimagestoenhancetherobustnessofthealgorithm.Intheprocessoffeatureextraction,wecombinetheWaveletPacketDecompose(WPD)with(2D)2PCA.First,theWPDisdidontheblockfaceimages;then,weextracttheintegretedimageswiththeoptimalnodeofsub-blocksbyusing(2D)2PCA.TochoosetheoptimalnodefromallnodesviaWPD,amethodthatcanselecttheoptimalnodeaccordingtotheimagerecognitionaccuracyofeachnodeisproposed.First,theimagerecognitionaccuracyofallnodesineachsub-block,whichhasexperiencedWPD,iscalculated;then,choosefivenodescorrespondingtothefirstfivebiggestrecognitionaccuracytocarryonweightedfusion.Ontheproblemofselectingweightvalue,aimagefusionmethodthatusestherecognitionaccuracyofeachnodedirectlyisproposedinthispaper.Whenwestudythevalidityofthealgorithm,wecomparethenewalgorithmproposedinthispaperwiththePCA、2DPCA、(2D)2PCA、WT-(2D)2PCAalgorithm.ExperimentalresultsontheORLfacedatabaseshowthattheproposedalgorithmhasahigherrecognitionaccuracythantheotherfourkindsofalgorithms.Whenwemakearesearchaboutthere