资源描述：
《基于GABOR小波的人脸识别的单样本问题研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、上海交通大学硕士学位论文基于Gabor小波的人脸识别的单样本问题研究姓名：姚荣华申请学位级别：硕士专业：计算机软件与理论指导教师：卢宏涛20070101上海交通大学硕士学位论文AbstractFacerecognitionisanimportantsubjectinartificialintelligencefield,andithasgainedextensiveattentionfromresearchersinthepastdecades.Duetothegoodcharacteristicofhumanfaceimage’s
2、Gaborwaveletfeature,facerecognitiontechnologybasedonGaborwaveletisaverypopularmethod.Facerecognitionisanintersectingsubject,andthispaperisfocusingonfeatureextractionoffacerecognition,andalsothispaperproposesanewsolutionforsinglesampleprobleminfacerecognition.Thispaperu
3、tilizesthepropertythatacomplexnumbercanberesolvedintomagnitudeandargumentinapolarcoordinatesystem,extractsargumentfeaturesfromhumanfaceimage’sGaborwaveletrepresentation,andgivesthedistributionfigureofhumanfaceimage’sGaborwaveletrepresentationcontainingargumentfeaturesa
4、ndmagnitudefeatures.Forthesinglesampleproblem,bymakinguseofmagnitudefeaturesandnewextractedargumentfeatures,thispaperproposesanovelEnrichedGaborfeaturebasedPrincipalComponentAnalysis(EGPCA)algorithm.BasedontheEGPCAalgorithm,thispaperimplementsafacerecognitionsystem,and
5、comparesEGPCAwith22(,PC)AE(PC)A,andSVDPerturbationinafacerecognitiontask.ExperimentalresultsonFERETfacedatabaseshowthatEGPCAcanachieve89.5%recognitionratewhenonlyonetrainingimageperpersonisavailableinfacerecognition,whichissuperiortootherthreealgorithms.Forthepurposeof
6、comparingtheefficiencyofargumentfeaturesandphasefeaturesinfacerecognition,thisthesisappliesthesetwokindsoffeaturesinfacerecognitionexperimentsbasedonFERETandORLfacedatabases.Experimentalresultsonbothtwofacedatabasesshowthatargumentfeaturesaresuperiortophasefeatures.Key
7、Words:FaceRecognition,GaborWavelet,FeatureExtraction,PrincipalComponentAnalysis,SingleSampleII上海交通大学硕士学位论文符号说明PCA:PrincipalComponentAnalysis主成分分析LDA:LinearDiscriminantAnalysis线性判别分析ICA:IndependentComponentAnalysis独立成分分析GA:GeneticAlgorithm遗传算法EFM:EnhancedFisherlineardis
8、criminantModels增强的Fisher线性判别模型FFT:FastFourierTransform快速傅立叶变换MSE:MeanSquareError最小化均方误差2(PC)A:Projection-CombinedPrin
