资源描述：
《Using Wikipedia to Translate OOV Terms on MLIR英文资料》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、INCORPORATESUPPORTVECTORMACHINESTOCONTENT-BASEDIMAGERETRIEVALWITHRELEVANTFEEDBACKPengyuHong,QiTian,ThomasS.HuangIFPGroup,BeckmanInstituteUniversityofIllinoisatUrbana-Champaign,Urbana,IL61801,USA{hong,qitian,huang}@ifp.uiuc.eduABSTRACTCurrently,[5,6]onlyusesthepositiveexamplesasfeedba
2、ck.TheinformationimpliedbythenegativeByusingrelevancefeedback[6],Content-BasedImageexamplesisneglected.Moreover,[5,6]requiretheuserRetrieval(CBIR)allowstheusertoretrieveimagestoprovidepreferenceweightsfortherelevantimages,interactively.Beginwithacoarsequery,theusercanwhichsometimesis
3、difficulttaskfortheuser.Anselectthemostrelevantimagesandprovideaweightofexampleofusingboththepositiveandnegativepreferenceforeachrelevantimagetorefinethequery.examples,whicharechosenbytheuser,forimageThehighlevelconceptbornebytheuserandperceptionretrievalcanbefoundinFourEyes[8].Thesy
4、stemlookssubjectivityoftheusercanbeautomaticallycapturedbyatallthelocalmodelsanddetermineswhichmodelorthesystemtosomedegree.Thispaperproposesancombinationofmodelsbestcoversthepositiveexamples,approachtoutilizebothpositiveandnegativefeedbackswhilesatisfyingtheconstraintsimpliedbythene
5、gativeforimageretrieval.SupportVectorMachines(SVM)isexamples.appliedtoclassifyingthepositiveandnegativeimages.Inthispaper,weproposetoapplySupportVectorTheSVMlearningresultsareusedtoupdatetheMachinetolearningpositiveandnegativefeedback.Thepreferenceweightsfortherelevantimages.Thislear
6、ningresultsarefurtherusedtohelpautomaticallyapproachreleasestheuserfrommanuallyprovidingdecidepreferenceweightsforthepositiveimages,whilepreferenceweightforeachpositiveexample.thewayofcalculatingqueryconceptremainthesameExperimentalresultsshowthattheproposedapproach[6].Therestofpaper
7、isorganizedasfollows.Beginninghasimprovementoverthepreviousapproach[5]thatwiththediscussionofrelevancefeedbacktechniqueinusespositiveexamplesonly.section2,webrieflydescribeSupportVectorMachines(SVM)insection3.TheapplicationofSVMtoCBIRis1.Introductionexplainedinsection4.Theproposedmet
8、hodsareteste
