基于Gabor紋理的AAM建模.pdf

1、西安電子科技大學碩士學位論文基于Gab紋理的AAM建模姓名：王宇申請學位級別：碩士專業(yè)：電路與系統(tǒng)指導教師：李潔20100101ABSTRACTABSTRACTThefeatureextractionandmodelingofthetargets，whicharethekeypointsoftheareaofcomputervision，haveimportantinfluencesaboutimageunderstandinganda

2、nalysisActiveAppearanceModels(AAM)，advancedbyETCootesetal，isaneffectiveandexactmodelingalgorithmoffeatureextractionThisalgorithmutilizesPrincipalComponentAnalysis(PCA)totrainandanalyzeimages，usingastatisticalwaytomodelbo

3、thshapeandtextureofobjectsAAMcalleffectivelywipeoffthenonlinearcouplingbetweenshapemodelandtexturemodel；reducethecomputationalcomplexityoffeaturesSo，AAMfeatureextractionplaysimportantrolesintargetdetectionrecognitionandt

4、rackingHowever，astheapplicationexpanding，thebasicAAMcannotsatisfytheneedsofaccuratenessInthispaperwedoaresearchabouttheAAMmodelingprocess，tryingtoimproveitformthreesides：AtfirstthetexturerepresentationofbasicAAMisverysim

5、pleInthispaperGaborfilters、而mgooddescriptionofscaleandorientationareintroducedtoextractamplertexturefeaturesoffacialimagesThesefeaturesCanhelpAAMgetmoreaccuratefacialfeaturelocationpointsTherefore，theAAMbasedonGaborfilte

6、rscanavailablyimprovethefeatureextractionabilityofbasicAAMSecondlybecausetheGabortexturefeaturesresultinthedecreaseofmodelingprocess，weutilizelocalGaborfiltersbanktepresenttheimagetextureinorderteducethecomputationalThen

7、，usingGeneticAlgorithm(GA)，weproposetOfindthebestGaborfiltersbanktoextractthetexturefeaturesFromthesemeasures，wereducethecomplexityofthemodelingprocessofGaborbasedAAMwithoutaffectingtheaccuratenessAtlastinordertoovercome

8、thedisadvantagesofvectormethod，weadvanceamodelingalgorithmbasedontensorGaborfeatureTraditionalvectormethodwhichdoesnotconsiderthespatialpositioninformationcurrentlymaydestroytheinnerdatastructureoftheoriginaldata；cannoto

9、btainacompactermodelTherefore，weintroduceTensorSubspaceAnalysis(TSA)whichcallreservethespatialpositioninformationofthedatatoimprovecalculationalefficiencyTheexperimentalresultsshowthatthetheoriesinthispaperhavegoodperfor