摘要
Theproblemofrecognizingnaturalscenes,suchaswater,smoke,fire,wind-blownvegetationandaflockofflyingbirds,isconsidered.Thesescenesexhibitthecharacteristicdynamicpattern,buthavestochasticextent.Theyarereferredtoasdynamictexture(DT).Inreality,thediversityofDTsondifferentviewpointsandscalesareverycommon,whichalsobringgreatdifficultytorecognizeDTs.Inthepreviousstudies,duetonoconsideringofthedeformableandtransientnatureofelementsinDT,themotionestimationmethodisbasedonbrightnessconstancyassumption,whichseeminappropriateforaggregateandcomplexmotions.Anovelmotionmodelbasedonrelativemotionintheneighborhoodoftwo-dimensionalmotionfieldsisproposed.Theestimationofnon-rigidmotionofDTsisbasedonthecontinuityequation,andthenthelocalvectordifference(LVD)isproposedtocharacterizeDTlocalrelativemotion.SpatiotemporalstatisticsoftheLVDsisusedastherepresentationofDTsequences.ExcellentperformancesofclassifyingallDTsinUCLAdatabasedemonstratethecapabilityoftheproposedmethodindescribingDT.
出版日期
2016年04月14日(中国期刊网平台首次上网日期,不代表论文的发表时间)