摘要
Storageandretrievalofmultimediaobjectshasbecomearequirementformanycontemporarysystems．Forexample，givenanimagedatabase，onemaywanttoretrieveallimagesthataresimilartoaqueryimage．Asmanycontent-basedretrievaltechniquesfordigitalimageryuseafeaturevectorapproachtorepresentimagecontents，itisdesirabletoreducethedimensionalityofthedata，whilstmaintainingasmuchofitsoriginalstructure．Severaldimensionalityreductiontechniquesareavailable．ThemostpopularoneisPCA，whichworkswellforstaticdatabases．Inthispaper，wepresentanovelschemeforperformingPCA-baseddimensionalityreductionindynamicdatabases．Insteadofusingtheentiredataset，weonlyrecomputethePCAtransformmatrixontheupdatingdataset．Notethatthesizeoftheupdatingdatasetisusuallymuchsmallerthanthatoftheentiredata，thistechniquemayreducethePCAcomputationtimecomplexitywithoutlosingexactness．Inaddition，theupdatestothedatabasearebasedontheexistingdimensionality-reduceddatavectorsratherthantheoriginalhigh-dimensionaldatavectors，whichmayrelievethesystemoverheadforthemanagementoftheoriginalhigh-dimensionaldata．
出版日期
2004年02月12日（中国期刊网平台首次上网日期，不代表论文的发表时间）
