图像搜索引擎的存储与索引技术研究与实现

华中科技大学硕士学位论文图像搜索引擎的存储与索引技术研究与实现姓名：康达祥申请学位级别：硕士专业：计算机应用技术指导教师：章勤20060430I(VisuAl&SemanTicImageSearchEngine)WebVASTVASTC-VASTLinuxVFSI/OJAVAJSPJNI16CSHELLVASTIIAbstractAlongwiththesuddengrowthoftheWorldWideWebaswellastherapiddevelopmentofmultimediatechnologyandthecommunicationtechnology,retrievingandbrowsingtheInternetmultimediainformationefficientlyandeffectivelyquicklybecomespeople'surgentdemand.Atpresenttheexistingimagesearchengines,mostlyusesthetextkeywordsbasedandthelinkinformationbasedtechnologytocompletetheimagesearchandtheimageretrieval,buttheyhasneverusedthevisualcontentinformationoftheimageitself.Sothesesystems’retrievalprecisionremainssomecertainlimit.Meanwhilesomecontent-basedimageretrievalsystemsthatareonlyusinginformationcontentofimagesforimageretrievalaredifficulttotacklethesemanticgap.Inresponsetotheseshortcomings,thestudyofWebimagessemanticcontentandvisualimageryassociatedwiththekeyfeatureoftheintegratedtechnologygoes,andwearetodevelopacombinationofvisualimageandsemanticsearchenginenamedVASTtoimprovetheaccuracyofWebsearchimages,andtoprovideinformationtousersindiverseways.VASTimagesearchengineindexingsystemforthestorageandvisualinformationofthenetworkimagesisanintegrationofsemanticinformationindexingandretrievalofimagesanddatamanagementoperationrequestedinterface.AdaptivefuzzymechanismisthebasedalgorithmoftheVASTimagesearchenginestorageandindexingsystem,andalsoisakeypointtobuildtheimageindextree.Fuzzyclusteringanddynamicmulti-levelindexingtreewithdynamic,flexible,non-dependenceonthecharacteristicsofthetargetsampledata,isagoodmechanismtoovercometheambiguityoftheclustercenterC-meansensitivityandthelocalizedclusteringofthenetworkcharacteristicsofthedistributionofimagedatatosolicithighladder,thestructurehasbetterbalancesymmetry.TheparallelstoragearchitectureoftheVASTimagesearchenginesindexingandstoragesystemisanotherresearchfocus.TheparallelframeworkdocumentadoptedLinuxphysicsinvestigatinginterfacesbetweensystemsandservicestoshieldtheactualphysicalstoragesystemoperation,whentheoperationofthesystem-issueddocumentimagesavailable,thesystemimagesearchsystemspecifictotheimplementationoftheIIIcoresystemresources.Theindexdatastructurethatcanbedistributedstorage,parallelprocessing,thiswouldincreaseI/Ospeed,thusspeedinguptheretrievalspeed.ThesystemusestheJAVArealizationnetworkreptiletobepartialandsystemthroughJNItechnologytransfercompleted16-dimensionalimageprocessingpackagesfeatureextractionandimageframeintegration.ParallelstructureestablishedbyclusteringstoragesystemandCcode,andsomecommoncodeShellachieve.Systemtestingshowedthatvastvisualimageswithframeaccuratesearchandretrievalsystemtothemorecommonimagessearchenginesrateincrease,retrievalperformance,goodtomeetuserneeds.Keywords:Searchengine;ImageSearchengine;Visualsemanticsfusiontechnology;Imageindexing;Fuzzyclustering;Content-basedimageretrieval_____111.1[1][2][3]WebVAST(VisuAl&SemanTicImageSearchEngine)WebWeb1.21.2.1[3-5](SearchEngine)InternetInternet2[6](robotspider)Internet[7,8][9][10,11]20063CNNIC1765.7%()1.2.2[12-15]31.2.3[16]()“”“”1.2.4[17](1)rPage-Rank4(2)(3)Web1.3.270WEBWEBWEBWEBWEB(1)GoogleGoogleWEB2WEB42.8(FullText)()WEBGoogleGoogle3.9WEB(2)AllTheWEB5Al1TheWEBFASTInternet.WEBFTPMacromediaFlashMP322521WEBWEB4500(3)BaiduBaiduBaidu2003MP3FlashBaidu1.3.390CBIR1)2)3)4)5)(1)QBICQBIC[18-20]IBMCBIR(RGB)(Yiq)(Lab)MTMkTamura(circularity)[21]KLTR*[22](2)VirageVirageQBICVirage[23]VirageQBICJeffrey6()()()GuptaJain9(3)PhotobookPhotobook[24]MITFourEyesPicard[25,26]“”[27,28](4)MarsMars[29]MARS(DBMS)(IR)MARSDBMSIR()()MARSMARS(5)BlobworldBlobworld[30,31]CaliforniaBerkley[32]iFind[33]WEBscope-CBIR[34]MiresSVM[35]71.3.4Webweb(1)ImageRoverImageRover[36,37](LatentSemanticIndexing)HTML(PrincipalComponentAnalysisPCA)k-dLmMinkowski[38][39](2)DiogenesDiogenes[40-43]webURLhttp//ranger.uta.edu/~alp/crawlersgoogleAltaVistaSnapURLsURL/HTML8(3)AtlasWISEAtlasWISE[44-47]Sherbrookewebgoogleyahoothumbnails()linkscolorhistograms[48,49]()edgeorientationhistograms[50]()(tagscaptionspagetitles(surroundingtext))(4)PicToSeekPicToSeek[51-55]URL()(correlationfunction)(histogramintersection)k(K-nearestneighbors)(query-pointmovement)(colordepth)(colorvariation)(colorsaturation)(colortransitionstrength)Fisher(Fisher'slineardiscriminatemothod)SR(5)AMOREAMORE[56-57]NECC&C[]WEB9(Similar)AMORE“”URLWebWeb-WISE[58]RIME[59]DrawSearch[60]91.4Web(1)(2)(3)(4)(1)(2)(3)(4)GoogleYahooBaidu1.5VASTWebWeb10VASTC-VASTI/OI/O1.6VAST(VisuAl&SemanTicImageSearchEngine)CVASTVASTVASTVASTVASTVASTVASTPortalVAST111.7“”122VASTVAST(VisuAl&SemanTicImageSearchEngine)VAST2.1VASTWEBHTML2.2VASTVASTVASTCBISVASTCBISCBISWEBWEBWEB13querybyimageexampleInternetWebW1(MA)W2(MA)Wn(MA)W1WnVASTHTMLHTML(VAST)14AgentAgentVASTVAST2.3VAST2.2WebAg