基于顏色的具有相關性反饋的圖像檢索系統(tǒng).pdf

1、浙江大學碩士學位論文基于顏色的具有相關性反饋的圖像檢索系統(tǒng)姓名：張守武申請學位級別：碩士專業(yè)：信號與信息處理指導教師：葉秀清2002.2.1ABSTRACTRecentvearshaveseenarapidincreaseoftilesizeofdigitalimagecollectionsEveryday，bothmilitaryandcivilianequipmentgeneratesgiga—bytesofimagesHugeam

2、ountofinformationisoutthereHoweverwecannotaccesstoormakeuseoftheinformationunlessitisorganizedsoastoallowefficientbrowsing，searchingandretrievalImageRetrievalhasbeenaveryactiveresearchareasincetheI970’s，withthethrustfrom

3、twomajorresearchcommunities，DatabaseManagementandComputerVisionThesetworesealchcommunitiesstudyImageRetrievalfromdifferentanglesonebeingtext—basedandtheothervisual。basedThetextbasedImageRetrievalcanbetracedbacktothelaIel

4、970‘sAverypopularframeworkofhnageRetrievalcontainsDataModeling，MultiDimensionalIndexing，QueryEvaluationetcHoweverthereexisttwomajordifficulties，especiallywhenthesizeofimagecollectionsislarge(tensorhundredsofthousands)One

5、isthevastamountoflaborrequiredinmanualimageannotationTheotherdifficultywhichismoreessential，resultsfromtilerichcontentintheimagesandtilesubjectivityofhumanperceptionThinis，forthesameimagecontent，differentpeoplemayperceiv

6、eitdifferentlyTheperceptionsubjectivityandannotationimpreciseflessmaycauseunrecoverablemismatchesinthelaterretrievaIprocessesIntheearly90’s，becauseoftheemergenceoflargescaleimagecollections，thetwodifficultiesfacedbythema

7、nualannotationapproachbecamemoreandmoreacuteToovercomethesedifficulties，contentbasedImageRetrieval(CBIR)wasproposedThatis，insteadofbeingmanuallyannotatedbytext—basedkeywords，imageswouldbeindexedbytheirownvisualcontent，su

8、chasColorTexture，Shape，etcSincethen，manytechniquesinthisresearchdirectionhavebeendevelopedandmanyImageRetrievalsystems，bothresearchandcommercial，havebeenbuiltTheadvancesinthisresearchdirectionaremainlycontributedbytheCom

9、puterVisioncommunityThisapproachhasestablishedageneralframeworkofImageRetrievalfromanewperspectiveHowevertherearestillmanyopenresearchissuestobesolvedbeforesuchretrievalsystemscanbeputintopracticeRegardingContentBasedIma

10、geRetrieval，wefeelthereisaneedtosurveywhathasbeenachievedinthepastfewyearsandwhatarethepotentialresearchdirectionswhichcanleadtocompellingapplicationsInthispaperwewilldevoteoureffortprimarilytotheContentbasedImageRetriev

11、alparadigmTherearefourfundamentalbasesforContentBasedImageRetrievalieVisualFeatureExtraction，MultiDimensiona】IndexingRelrievalSystemDesignandRelevanceFeedbackIreviewvariousvisualfeaturesandtheircorrespondingrepresentatio

12、nandmatchingtechniquesTofacilitatefastsearchinlargescaleimagecollections，effectiveindexingtechniquesneedtobeexploredIverealizedtileCBIRbased011ColorfeatureswithRelevanceFeedbackandgivenSOlneconclusionanddiscussTheCBIRsys