摘要
Inmanyvideostandardsbasedoninter-framecompressionsuchasH.26xandMPEG,block-matchingalgorithmhasbeenwidelyadoptedasthemethodformotionestimationbecauseofitssimplicityandeffectiveness.Nevertheless,sincemotionestimationisverycomplexincomputing.Fastalgorithmformotionestimationhasalwaysbeenanimportantandattractivetopicinvideocompression.Fromtheviewpointofmakingmotionvectorfieldsmoother,thispaperproposesanewalgorithmSMVFAST.Onthebasisofmotioncorrelation,itpredictsthestartingpointbyneighboringmotionvectorsaccordingtotheirSADs.Adaptivesearchmodesareusedinitssearchprocessthroughsimplyclassifyingmotionactivity.AfterdiscoveringtheubiquitousratiobetweentheSADsofthecollocatedblocksintheconsecutiveframes,thepaperproposesaneffectivehalf-stopcriterionthatcanquicklystopthesearchprocesswithgoodenoughresults.ExperimentsshowthatSMVFASTobtainsalmostthesameresultsasthefullsearchatverylowcomputationcost,andoutperformsMVFASTandPMVFASTinspeedandquality,whichareadoptedbyMPEG-4.
出版日期
2003年01月11日(中国期刊网平台首次上网日期,不代表论文的发表时间)