搜索
doi10．3969/j．issn．1002－0802．2014．01．007Gabor361021在人脸表情识别中，针对Gabor小波变换特征维数很大的问题，提出了一种新的多方向特征编码方法。通过对Gabor特征幅值进行统计处理，将每个像素点同一尺度不同方向的Gabor特征幅值阈值化成二进制，加强了Gabor小波对图像局部结构信息的表征。同时，结合了类似旋转不变LBP的方法对图像进行降维。为了进一步提高表情的正确识别率，采用一种局部区域融合的方法，最后在JAFFE表情库上进行测试，得到比较好的识别率，验证了所提方法的有效性。表情识别Gabor小波特征降维局部区域融合TP391A1002－0802201401－0033－04FacialExpressionＲecognitionbasedonGaborWaveletTransformXIEJin－fang1FENGGui2CollegeofInformationScienceandEngineeringHuaqiaoUniversityXiamenFujian361021ChinaAbstractAimingatthehighdimensionsofGaborwaveletsfeatureinfacialexpressionrecognitionanewmethodforcodingGabormulti－orientationfeaturesisproposed．AfterstatisticalprocessingoftheGaborfeaturesandbinaryprocessingofthemagnitudesofdifferentorientationsthelocalstructuralinformationrepresentationoftheimagebyGaborwaveletisenhanced．MeanwhileforreducingtheimagedimensionsasimilarmethodasintheＲotationInvariantLBPisusedanddiscussed．Furthermoreamethodforfusingthelocalfacialregionsisemployedthustoimprovetheexpressionrecognitionrate．TheexperimentonJAFFEdatabaseindicatesafairlygoodrecognitionrateandthisverifiestheeffectivenessoftheproposedalgorithm．KeywordsfacialexpressionrecognitionTTTdimensionreductionofGaborwaveletsfeaturefusionoflo-calregions0、、、、1。。。Gabor、2。Gabor。3fastPCA+LDA3Gabor4Gabor5－6。GaborGabor。7。JAFFE·33·47120141CommunicationsTechnologyVol．47No．1Jan．2014。11．1Gabor2D－Gabor2。2D－Gabor8ψuvz=‖kuv‖2σ2·exp‖kuv‖2‖z‖22σ2·expikuvz－exp－σ22[]1kuv=kvcosφusinφuTkv=2－v+22πGaborφu=uπKz。σGabor2π。Iz58Gaborψuvzu=01…4v=01…7K=840Gabor。IzGaborGabor1。1GaborFig．1Gaborwaveletfeaturesoffacialexpressionimage1．2Gabor。。18Gmeanz=∑7v=0Guvz82Puvz=1Guvz＞GmeanzPuvz=0Guvz≤Gmeanz{3Gmeanzz=xy8GuvzzuvGaborzPuvzv∈01…7u∈01…4。Gabor。28BGP=∑7v=0Puvz×2v43Gabor4BGP256LBP9BGPriBGP36。52。2Fig．2Codedfeatureimagesondifferentscales。10。32×26。1．311FeyeFmouthFface。Feature=FfaceFeyeF{}mouth。·43·．txjszz．com201422．1JAFFE10213、、、、、73～4。70143。2．23。3Fig．3Flowchartofexpressionrecognitionsystem、128×104、32×104、32×564。4Fig．4PreprocessedimageGabor128×104×402880Gabor180。PCA+LDA3k。2．31face1。1Table1Experimentalresultsoffacialregions/%91．693．790．993．31009593．393．98Gabor。。2face、eye+mouthface+eye+mouth2。2Table2Experimentalresultsofthefaceandlocalregions/%/sface93．91．27eye+mouth93．20．53face+eye+mouth94．91．812face+eye+mouthfaceeye+mouthface。33。3Table3Comparisonoftheproposedmethodwithothermethods/%/sGabor+2DPCA91．21．34Gabor+FastPCA+LDA385．721．3294．91．81·53·47Gabor13Gabor。①GaborGaborBGPriGabor②。。1．J．2011380425－30．JIANGBinJIAKe－binYANGGuo－sheng．ＲesearchAdvanceofFacialExpressionＲecognitionJ．ComputerScience2011380425－30．2．GaborJ．201227081277－1280．KONGＲuiZHANGBing．DesignofGaborFilters’Param-eterJ．ControlandDecision201227081277－1280．3．GaborJ．2012200363－66．WANGFu－longBOHua．FacialExpressionＲecognitionbasedonGaborWaveletTransformJ．ElectronicDesignEngineering2012200363－66．4．GaborFisherSVMJ．201128041536－1543．HUANGYong－mingZHANGGuo－baoDONGFeietal．MultipleFeaturesExtractionUsingGaborWaveletTransformationFisherFacesandIntegratedSVMwithApplicationtoFacialExpressionＲecognitionJ．Appli-cationＲesearchofComputers201128041536－1543．5LAJEVAＲDISMHUSSAINZM．ANovelGaborFilterSelectionbasedonSpectralDifferenceandMinimumEr-rorＲateforFacialExpressionＲecognitionC//DigitalImageComputingTechniquesandApplicationsDIC-TA2010InternationalConference．SydneyNSWIEEE2010137－140．6．GaborJ．2010330797－101．YINYongLIＲong－gangWANGJian－dongetal．AFuzzyFusionApproachtoGabortransformOptimalChan-nelsforFacialExpressionＲecognitionJ．JournalofChongqingUniversity2010330797－101．7．J．20092506963－966．YINGZi－luZHANGYou－weiLIJing－wen．FacialExpressionbyFusingLocalFacialＲegionsJ．SignalProcessing20092506963－966．8．GaborJ．20084112259－261．ZHAOSan－pingSULi－qiu．GaborTransformationbasedImageTextureＲecognitionJ．CommunicationsTechnology20084102259－261．9ZHANGLZHOUZLIH．BinaryGaborPatternAnEffi-cientandＲobustDescriptorforTextureClassificationC//ImageProcessingICIP201219thIEEEInterna-tionalConferenceon．OrlandoFLIEEE201281－84．10．GaborMVBoostJ．200744071089－1096．LIUXiao－minZHANGYu－jin．FacialExpressionＲecognitionbasedonGaborHistogramFeatureandMV-BoostJ．JournalofComputerＲesearchandDevelop-ment200744071089－1096．11KHANＲAMEYEＲAKONIKHetal．ExploringHumanVisualSystemStudytoAidtheDevelopmentofAutomaticFacialExpressionＲecognitionFrameworkC//ComputerVisionandPatternＲecognitionWork-shopsCVPＲW2012IEEEComputerSocietyConfer-ence．ProvidenceＲIIEEE201249－54．1988—XIEJin－fang1988－femalegraduatestudentmajoringinpatternrecognition．1960—、、、、。FENGGui1960－femalePh．D．professormainlyworkingatsignalandinformationprocessingcommunicationsys-temsandinformationsecuritytheorymultimediacommunica-tioncomputervisionpatternrecognitionandimageunder-standingetal．·63·．txjszz．com2014