为什麼image无法处理的像text一样好

《为什麼image无法处理的像text一样好》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、Content-BasedImageRetrieval-ApproachesandTrendsoftheNewAgeRitendraDatta,JiaLi,andJamesZ.WangThePennsylvaniaStateUniversityMIR2005INTRODUCTION為什麼image無法處理的像text一樣好Textisman’screation,imagesareamerereplicaofwhatmanhasseenInterpretationofwhatweseeishardtocharacterizevisualsimilarity!=

2、semanticsimilarityCBIRhasgrowntremendouslyafter2000,notjustintermsofsize,butalsointhenumberofnewdirectionsexploredINTRODUCTIONINTRODUCTIONThetheoreticalfoundationbehindhowwehumansinterpretimagesisstillanopenproblemAbriefscanningofabout300relevantpaperspublishedinthelastfiveyearsrev

3、ealedthatlessthan20%wereconcernedwithapplicationsorreal-worldsystemsCBIR領域研究方向FeatureExtractionApproachestoRetrievalAnnotationandConceptDetectionRelevanceFeedbackandLearningHardwareandInterfaceSupportFeatureExtraction如何抽ColorFeature“AnEfficientColorRepresentationforImageRetrieval”(

4、比傳統histograms好)“MultiresolutionHistogramsandTheirUseforRecognition”(用在texturedimage)“ImageretrievalusingcolorhistogramsgeneratedbyGaussmixturevectorquantization”(利用GMVQ抽colorhistogram)FeatureExtractionColor+Texture抽取“Wavelet-BasedTextureRetrievalUsingGeneralizedGaussianDensityandKu

5、llback-LeiblerDistance”Shape“ShapeMatchingandObjectRecognitionUsingShapeContexts”(isfairlycompactyetrobusttoanumberofgeometrictransformations)FeatureExtractionSegmentation“NormalizedCutsandImageSegmentation”(最重要的方向之一)“Blobworld:ImageSegmentationUsingExpectation-maximizationandItsAp

6、plicationtoImageQuerying”(我之前用過的方法)“SegmentationofBrainMRImagesThroughaHiddenMarkovRandomFieldModelandtheExpectation-MaximizationAlgorithm”(處理medicalimaging)FeatureExtraction線條相似度“Imageretrievalusingwavelet-basedsalientpoints”如何選擇featureApplication-specificfeaturesets(最直觀的)“SIMPLIc

7、ity:Semantics-SensitiveIntegratedMatchingforPictureLibraries”(semantics-sensitivefeatureselection)“FeatureSelectionforSVMs”(用classifier)ApproachestoRetrievalRegionbasedimageretrieval“AScalableIntegratedRegion-BasedImageRetrievalSystem”region-basedquerying(BlobWorld)Vectorquantizati

8、on(VQ)onimageblocks“Keyblo