像检索方法及其在图像数据库中应用

上海交通大学硕士学位论文彩色图像检索方法及其在图像数据库中应用姓名：潘红艳申请学位级别：硕士专业：控制理论与控制工程指导教师：敬忠良20070201CBIR(Content-BasedImageRetrieval)(1)SCMeanShiftHSVSCSpatialColorFeatureSpaceMeanshifthue(2)(3)Tif/TiflibtifEngineFirebirdDelphiInterbaseMATLABMeanshiftLIBTIF,DATABASETECHNOLOGYOFCOLORIMAGESABSTRACTContent-BasedImageRetrieval(CBIR)isusedtofindoutthetargetimagefromtheimagedatabaseaccordingtothegivenimagefeatures.Theimagefeaturescanbeextractedfromthesampleimagesprovidedorinputtedbycustomers.CBIRmainlycontainsfourkeytechniques,whichareimagedatabase,contentdescription,featureextractionandmatchingandfastsearching.Thisthesisdealswiththefeatureextractionandcomparisonofcolorfulnaturalimages.Itcontainsthreeaspects,whichareshapefeatureextraction,texturefeatureextractionanddesignandrealizationofaremotesensingimagesdatabaseonDelphi.(1)Shapefeaturedescribestheedgecharacteristicsofimageorpartofimageandthefeaturevectorisonaboundontheedgewhilefluctuatesmoothlyinsideit.Thispaperproposedanewmethodcombiningtheimprovedmeanshiftalgorithmandregionmethod.Themainachievementsandcontributionsaboutmethodsandalgorithmsaredescribedasfollows:SelectedtheSpatialcolorfeaturespaceinHSVspace;improvedtheweightsselectionandwindowselectionalgorithmofmeanshift;Theself-adaptivemethodhandleswiththeanisotropyofspaceandtheassignmentofweightsaccordingtodifferentreliability.Thenwefusesthespaceandcolorinformationbycombiningtheimprovedmeanshiftandregionmethod,onthebaseofwhichtheEuclideandistanceofthesevenofhueisappliedtorealizeretrieval.(2)Texturefeatureisoneoftheattributesofimage,whichdescribesthespacedistributionofgraylevelsofimagepixels.Animagecontainstextureiftheobjectsintheimagehaveadistinctbutnotsimplehuechange.TexturefeatureextractioninthisthesisisbasedonwavelettransformationandCo-occurrencematrix.Thecontributionsofthisalgorithm:thewavelettransformationgettheinformationfromdifferentfrequencybandsandthentheCo-occurrencematrixgettheirfeaturerespectively.FinallytheEuclideanDistanceisusedtomeasurethesimilaritybetweendifferentimages.(3)ImageDatabaseAimingatcomplexTIF/TIFFformatremotesensingsuperimages,wedesignandrealizetheminiatureextractionbasedimagedatabaseusinglibtiffunctionlibrarybyconnecttheDelphiandInterbaseviafirebirddatabaseengine.Thecontributionisasfollows:linkingtheoriginalimageandminiaturethroughforeignkey;dynamicdisplayoftheoriginalimages;modulebasedarchitectdesignincreasetheextensibilityandreusabilitysoastofacilitatethemanagement;improvedtheimageprocessingmethodofprogrambyacceleratethespeedbymorethantwentytimes.Inthisthesis,thetypicalmethodsonfeatureextractionofnaturalcolorimagesareintroduced,twonewmethodsaredesignedtoextractthetextureandshapefeaturesofnaturalcolorimages,andthecomparisonofthefeaturevectorsoftwonaturalcolorimagesisproposed.Finally,thesimulationofthethreemethodsusingMATLABachievesgoodresults.KEYWORDS:Self-AdaptiveMeanshifttexturefeatureshapefeatureLIBTIFFast-MiniatureExtraction..200722620072262007226411ASIDCBIRContent-BasedImageRetrieval[1]CBIRCBIRJainRamesh[2]VIMS(VisionInformationManagementSystem)AlmadenIBMQBIC(QuerybyImageContent)VirageVirageExcaliburWareRetrievalMITPhotobookUIUC)SystemRetrievalandAnalysisMultimedia(MARSMIRESInternetRWebscopeCBEnginePhotoNavigatorPhoto51Virage***2QBIC****3Photobook**4Piction***5Chabot**1.112121(l)(2)(3)QBICMethre[3][4](DependentScalarQuantization)(DSQ)6StrickerDimai[5]ChangSmith[6](SequentialLabelingAlgorithm)(Back-Projection)Sarkar[7](Clustering)C[8]CarevicK[9]3C[10]CBeni-Xiesugno-FukayamaSC[11][12]MeanShift[13]ChengMeanshiftBerkeley[14]C[15]MeanshiftXIAN-JIUGUO[16]7K.SinghManeesh[17]CHENPAN[18]ManhanttanEuclidean(HistogramIntersection)[19]TomasiGuibasRubnerHausdorffHausdorff[20]Euclidean12280Kashyap91F.S.Cohen(GMRGaussianMarkovRandomField)999100D.K.PanjwaniG.HealyGMRFM.R.TurnerGaborA.C.BoviliA.K.JainT.R.ReedM.UnserGahorCarterMorletMexcan698A.Lame8J.Fan123()(ShapebyVolume)Hough[21][22]Patch-typePart-type[23],[24][25]R.LiuH.Zhang3D[26]A.P.ManganR.T.Whitaker3D[27]D.L.Page3D[28][29](DeformableModel)(ElasticDeformableModel)1973Widrow(ActiveContourModelSnake)M.Kass19889(parametricdeformablemode)(parameterizedcurses)(parametricmapping)GrenanderAmitCootes(activeshapemodel)(ShapeFactor)()Jagadsh(Point-AccessMethod)IBMQBICPAM13CBIC12CBIR10314[30]1GTif/Tiff1.1Fig1.1ArchitectureofRemoteSensingImagesROI111.5Tif/TiffAdobeLibtif1221:::22[31]H(f,i)ff[32]()()()13[33]ColorCoherenceVectorColorCorrelogram[34]:HSVLabHSV;();;23()(FourierShapeDescriptors)[35](Momentlnvariants)[36](FiniteElementMethod)[37][38][39][40][41][42]14231(1)(2)(3)8232233Kullback-LeiblerDistence(KLD)23415235(token)(curvature)N1ii}p{=1kkp,p+kmkm(orientation)kθ),(mkkθNN1kkk},m{)c(T==θjiττ(2.1)α[0,1])(norientatio)(norientatio),(d)(curvature)(curvature),(d),(d)1(),(d),(djijinorientatiojijicurvaturejinorientatiojicurvaturejiττττττττττατταττ−=−=−+=(2.1)AB(2.2)()11(,)min{(,)}nipiPiDABdnττ==∑(2.2)(M-tree)236(PAMPoint-AccessMethod)IBMQBIC16()2372D1DXY1Df(r)(r,f(r))XYXYUVUXVYuiv+(X,Y)N:()()(),0,1,1skukjvkkN=+=−L()sk101()()exp(2/),0,1,1NkswskjwkNwNNπ===−=−∑L()sw10()()exp(2/