基于用戶反饋和改進詞袋模型的圖像檢索.pdf

1、聲明本學位論文是我在導師的指導下取得的研究成果，盡我所知，在本學位論文中，除了加以標注和致謝的部分外，不包含其他人已經(jīng)發(fā)表或公布過的研究成果，也不包含我為獲得任何教育機構(gòu)的學位或?qū)W歷而使用過的材料。與我一同工作的同事對本學位論文做出的貢獻均已在論文中作了明確的說明。研究生簽名：雞盔彬年弓月竅日學位論文使用授權(quán)聲明南京理工大學有權(quán)保存本學位論文的電子和紙質(zhì)文檔，可以借閱或上網(wǎng)公布本學位論文的部分或全部內(nèi)容，可以向有關(guān)部門或機構(gòu)送交并授權(quán)其

2、保存、借閱或上網(wǎng)公布本學位論文的部分或全部內(nèi)容。對于保密論文，按保密的有關(guān)規(guī)定和程序處理。研究生簽名：3撻例r年弓月哆日英文摘要碩士學位論文AbstractWitllmerapiddeVelopmentofIntemetaJldmultimediatechnologies，suchaLsd追italc鋤eraSandimageeditiOntools，廿1eamountofimagesOnmeImemethasbeenexplosiVe

3、Howtosearchimagesaccordingtouser’squeryinSucha1arged撕baSef瓠tande伍cientlyhaLsbecomeameaIling血la11dchaJlengingproblemConsideringtlle1i血toftextbaSedimageretrieValsystem，whjchismainJybasedontextkeywordsaIldmanua】aImotation，c

4、omembaLsedimageretrievalhaScaughtresearchers’eyesbecauseofitsdirecUyusageofimagecontent，especiallyt11euseoflocaldescriptionHoweVer，也ereeXistsawell—l(110wnsemanticgapbe似eenthe10wleVeldescriptiona11dlli曲1eVeliIllagesemanti

5、c，whererelevancefeedbackcanbeabridgeBagofvisualwordmodelisaclaSsicmodelusedinmanycontem_basedimageretrievalSystemsT11ismodeltreatsanimageasac01lectionof“words“aIldusessoInemanJretechIliquesintextbaLsedretrieValThou曲theBo

6、VWmodelgainsagreatsuccessbecauseofitssimple，thereremainssomeproblemscausedbythesemantic—g印betweentlle10wleveldescriptionandmgh1eVelimagesemalltic，WherereleVallcefeedbackcanbeabridgeReleVancefeedbackaddsmeuserintot11eretr

7、ieVallo叩，so廿lesyStemcaIluse也eimagestheuserselectedaLsreleValltori玎eleVanttoleam廿1euserintentandretumbetterimagesmatsuittheuserimentThispaperstudiestechniquesinV01VedinimageretrieValsyStem，especiaJlytheBOVWmodelaIldreleva

8、ncefeedbackThisp印erintroducesa11interestregionbasedrelevallcefeedbackapproach廿1ataI】【alyzestheuserintentaccordingtotllereleVantimagesThe印proachusesRANSACtoproVideaspatiaJadjustingso2Lstole鋤t11einterestedregionThispaperal

9、sointroducesap眥ishs扛ategythatproVides10ngte冊optimizationforthesystemaccordingtomei玎eleVantimagesExtensiVeexp嘶mentsshowt11attherelevantfeedback印proachcanboosttheperfoHnallce謝tll廿1is印proachAccordingtot11isrelevancefeedback

10、S訂ategy，t11ispaperadjuststhefeaturewei曲ting印proachinBoVWmodelt0makeitbetterrepresenttheimageFhlally，basedontlleseresearches，tllispaperpresentsaJlimageretrieValsyStemdesignedfortheuserWh0hasclearsearchintentKeyword：image蕾