基于FPGA的人脸识别项目.ppt

《基于FPGA的人脸识别项目.ppt》由会员上传分享，免费在线阅读，更多相关内容在PPT专区-天天文库。

1、FacialExpressionRecognitionBasedonfieldProgrammableGateArrayJzau-ShengLin,Shao-HanLiou,Wu-ChihHsieh,Yu-YiLiao,HongChaoWang*,andQingHuaLan*DepartmentofComputerScienceandInformationEngineeringNationalChin-YiUniversityofTechnology*EmbeddedLaboratory,HuaxiaVocationCollege,Xiamen,China1Outline

2、5.Experimentalresults4.HardwareImplementation3.PreprocessingforFFE2.DWT-CMAC1.Introduction6.Conclusion21.IntroductionTheCMAC(JamesAlbus)non-learningmappinglocalgeneralizationlowcomputationcomplexityrapidconvergencespeedfastpracticalhardwarerealizationTheproposedhardwareimplementationofare

3、cognitiontechniquewithFPGAThe2DDWTovertheentirefaceimageasafacialexpressionfeatureextractorTheCMACwithclusteringmemoryasaclassifier.32.DWT-CMAC(1/3)Differenceimage=givenexpressionimage–neutralimageHaarwavelettransformspacedomainfrequencydomainTheimageisdividedintofournon-overlappingmulti

4、-resolutionsub-bandsCMACwithCMAnS-bitinputvectorisencodedtoquantizeintocclusterswhichincludingbbitsTheoutputlayerisk-layerparallelmemoriestomemorizekfacialexpressions42.DWT-CMAC(2/3)TheproposedCMACTheweightcanbeadjustedwiththesteepingdescentruleasfollows:52.DWT-CMAC(3/3)Architectureofthef

5、acialexpressionrecognitionofCMAC63.PreprocessingforFFE(1/2)73.PreprocessingforFFE(2/2)84.HardwareImplementationBlockRAMSRAMVGADWTIPCMACIPRS23294.1DWTIP(1/2)ControllermoduleMemoryofbaseaddress=Row×128＋ColumnDWTModuleOperationmoduleHaarwaveletTransformationDWT_DONE=HIGH104.1DWTIP(2/2)Blockd

6、iagramoftheDWTmodule114.2CMACIP(1/4)ThetrainingphaseTheinputdatalow-frequencyband(32×32bytes)ofHaarDWTtransformationdivideanun-overlap2×2blockasatrainingsample(tosimplifytheinputcodingcircuitoftheCMAC)SRAMTostorethecodingaddresseswithasizeof16×16×12=3072bits.124.2CMACIP(2/4)CMACTrainingc

7、ircuit134.2CMACIP(3/4)TherecognitionphaseTheunknowninputdatamemoryTheWeightSummationModuletofindtheweightsTosumthemastotalweightstoexpresstheintensitiesfordifferentfacialexpressions.TousetheMaxmoduletofindthemaximumfacial-expressionintensity.144.2CMACIP(4/4)Blockdi