資源描述：
《2011-Bilinear deep learning for image classification》由會員上傳分享，免費在線閱讀，更多相關內容在學術論文-天天文庫。
1、BilinearDeepLearningforImageClassificationSheng-huaZhongYanLiuYangLiuDepartmentofComputingDepartmentofComputingDepartmentofComputingTheHongKongPolytechnicUniversityTheHongKongPolytechnicUniversityTheHongKongPolytechnicUniversityHungHom,KowloonHungHom,KowloonHungHom
2、,Kowloon999077HongKong,P.R.China999077HongKong,P.R.China999077HongKong,P.R.Chinacsshzhong@comp.polyu.edu.hkcsyliu@comp.polyu.edu.hkcsygliu@comp.polyu.edu.hkABSTRACTmethodscanberoughlydividedintotwobroadfamiliesofapproaches:parametricandnonparametricclassifiers.Para
3、metricImageclassificationisawell-knownclassicalprobleminclassifiers,alsoknownaslearning-basedclassifiers,requireanmultimediacontentanalysis.Thispaperproposesanoveldeepintensivetrainingphaseoftheclassifierparameters(e.g.,thelearningmodelcalledbilineardeepbeliefnetwo
4、rk(BDBN)forparametersofSVM[2],Boosting[3],fragmentsandobjectpartsimageclassification.Unlikepreviousimageclassificationmodels,[4],decisiontrees[5],webgraphs[6],hierarchicalclassificationBDBNaimstoprovidehuman-likejudgmentbyreferencingthemodels[7],etc.).Todate,thelea
5、dingimageclassifiersarearchitectureofthehumanvisualsystemandtheprocedureofparametricclassifiers,particularlySVM-basedmethods.intelligentperception.Therefore,themulti-layerstructureoftheNonparametricclassifiersmaketheirclassificationdecisionscortexandthepropagationo
6、finformationinthevisualareasofdirectlyonthedata,andrequirenotrainingofparameters[8].thebrainarerealizedfaithfully.UnlikemostexistingdeepRecently,intheliteratureonmultimedia,manypapersfocusedonmodels,BDBNutilizesabilineardiscriminantstrategytothespecificapplications
7、;forinstance,landmarkimagesimulatethe“initialguess”inhumanobjectrecognition,andatclassification[9],sportsgenre&viewtypeclassification[10],agethesametimetoavoidfallingintoabadlocaloptimum.Toimagesclassification[11]andaffectiveimagesclassification[12]preservethenatur
8、altensorstructureoftheimagedata,anovel[13].Inaddition,camerametadataareutilizedforclassificationdeeparchitecturewithgreedylayer-wisereconstructio
