基于感興趣區(qū)域的圖像檢索方法.pdf

1、中國(guó)科學(xué)院計(jì)算技術(shù)研究所碩士學(xué)位論文基于感興趣區(qū)域的圖像檢索方法姓名：斯白露申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用指導(dǎo)教師：高文2002.6.1ImageRetrievalMethodBasedonRegionsofInterestSIBailu(ComputerApplication)Directedby：GAOWen(Professor)AbstractWiththedevelopmentofdigitalmediatechnology

2、，moreandmoredigitalimagesareproducedHowtoorganize，store，represent，queryandretrievalthesemagesisanurgentproblemTheContentBasedImageRetrieval(CBIR)providesahelpfulmethodforthisgoalTherearethreekeyissuesinCBIR：1)howtepresen

3、tthecontentofanimage，2)howtomeasurethesimilaritybetweentwoimagesbasedontheserepresentationsand3)howtomaketheretrievalsystemlearnhumanbeings’subjectivemannersinimageunderstandingandcomparisonThisthesisdiscussesthefirsttwo

4、keyissuesofCBIR，focusingonimagecontentrepresentationandretrievalmethodsbasedonregionsofinterest(ROI)First，thecurrentfourimageretrievalmethodsarecomparedTheyareglobalfeaturesearch，colorlayoutsearch，region—basedsearchandth

5、enewlyappearedsearchmethodbasedonregions—ofinterestThen，afterthecurrentROIbasedmethodsarediscussed，anewimagerepresentationmethodisproposedThismethodusesavisualattentionmodeltoextractfocusesofattentioninanimageandmakesthe

6、masinterestpointsRegions—ofinterestaretheresultofinterestpointsclusteringTheimageisthusrepresentedbyasetofregions—ofinterest，whicharemoreperceptuallyimportantthanotherpaRRsoftheimageTocomparethesimilarityoftwoimages，anim

7、agecomparisonmethodbasedonregions—ofinterestwasproposedinthisthesisThismethodchoosesthemostsimilarregionsfromtheimagesasmatchedregions，andcomparethevisualsimilarityoftheimagesbycontentsimilarityandintegrityThelayoutsimil

8、arity，whichisinvarianttoimagerotationandflip，isalsocomparedaccordingtothepositionrelationshipofmatchedregionsFinally，webuildaprototypesystemtotesttheperformanceoftheproposedmethodKeyWords：imageretrieval，regionsofinterest