Fast Mode Decision Algorithm for Scalable Video Coding Using Bayesian Theorem Detection and Markov Process
IEEETRANSACTIONSONCIRCUITSANDSYSTEMSFORVIDEOTECHNOLOGY,VOL.20,NO.4,APRIL2010563
FastModeDecisionAlgorithmforScalableVideoCodingUsingBayesianTheoremDetection
andMarkovProcess
Chia-HungYeh,Member,IEEE,Kai-JieFan,Mei-JuanChen,Member,IEEE,
andGwo-LongLi,StudentMember,IEEE
Abstract—Thenewestvideocodingstandardcalledscalablevideocoding(SVC)providesbroadapplicationsinmultimediacommunications.SVCencoderconsumesgreatcomputationalcomplexitywhencomparedtopreviousvideocodingstandards.ThispaperpresentsafastmodedecisionalgorithmthatspeedsuptheSVCencodingprocessthroughprobabilisticanalysis.Themodeoftheenhancementlayeris rstpredictedbystatisti-calanalysis.Afterward,Bayesiantheoremisutilizedtodetectwhetherthepredictionmodeofthecurrentmacroblockisthebestornot.Themodeisfurtherpredictedandre nedbytheMarkovprocess.Experimentalresultsshowthattheproposedalgorithmsigni cantlyreducescomputationalcomplexitywithnegligiblepeaksignal-to-noiseratiodegradationandbitrateincreaseintheenhancementlayers.
IndexTerms—Bayesian,coarsegranularscalability,Markov,modedecision,scalablevideocoding.
I.Introduction
T
HEDEVELOPMENTandstandardizationofadvancedvideocodinghasattractedmuchattentionoverthepastdecade.Videocodingstandardsaredesignedforthepurposesofstorageandtransmission.Scalablevideocoding(SVC),anextendedversionofH.264/advancedvideocoding(AVC),hasbeen nalizedin2007[1],[2].SVCstandardizationaimstoencodeahigh-qualityvideothatconsistsofoneormoresubsetvideos.
TheprocessofdecodingsubsetvideosissimilartothatoftheexistingH.264/MPEG-4AVC.Thecomplexity,recon-structionquality,andquantityofdatainasubsetbitstreamarealsosimilartothatofH.264/AVC.SVChasseveralpossiblescalabilitymodalities,includingtemporal,spatial,SNR/quality/ delityscalabilities,andacombinedscalabilitythatintegratesthepreviousthreemodalities.Asubsetvideocouldbeeitheralowerspatial,temporal,orqualityversion
ManuscriptreceivedDecember13,2008;revisedApril7,2009andAugust9,2009.FirstversionpublishedJanuary29,2010;currentversionpublishedApril2,2010.ThispaperwasrecommendedbyAssociateEditor,V.Bottreau.C.-H.YehiswiththeDepartmentofElectricalEngineering,NationalSunYat-SenUniversity(NSYSU),Kaohsiung80424,Taiwan(e-mail:yeh@mail.ee.nsysu.edu.tw).
K.-J.Fan,M.-J.Chen,andG.-L.LiarewiththeDepartmentofElectricalEngineering,NationalDong-HwaUniversity,Hualien97401,Taiwan(e-mail:d9523005@ems.ndhu.edu.tw;cmj@mail.ndhu.edu.tw;d9323004@em93.ndhu.edu.tw).
Colorversionsofoneormoreofthe guresinthispaperareavailableonlineatdoc.guandang.net.
DigitalObjectIdenti er10.1109/TCSVT.2010.2041825
(oracombinationof,atleast,anyofthetwo)whencomparedtothesubsetvideoderivedfrom.Therefore,SVCsupportsvariouskindsofdevices,withdifferentrequiredbitratesordisplaycapacities.Inordertosupporttemporalscalability,SVCadoptshierarchicalBframeprediction.InthehierarchicalBstructure,allBframescanuseforwardandbackwardframesforprediction.Comparedtopreviouscodingstructures,suchasIPPPPandIBBP,thegainofpeaksignal-to-noiseratio(PSNR)inSVCisabout1dBhigherthanthatofH.264/AVC[2].HierarchicalBframepredictionnotonlyprovidesvariousframerates,butalsosigni cantlyimprovescodingef ciency.InSVC,amultilayervideocodingstructurecalledpyramidcodingschemeisadopted.InordertoimprovethecodingperformanceofSVC,itisnecessarytoremoveredundantdatagatheredamongdifferentsubsetvideosinthemultilayercodingstructure.SVCfurthersupportsthreenewinter-layerpredictionmodestoimprovetherate-distortionef ciencyoftheenhancementlayers(ELs)fortheELencodingprocess.Theyareinter-layermotionprediction,inter-layerresidualprediction,andinter-layerintraprediction.Althoughthenewcodingtoolsimprovecodingperformance,ittakeshighcom-putationalcomplexity.SVCinheritsthefeatureofH.264/AVC,whichsupportssevenblocksizesforinter-frameprediction.Amacroblock(MB)canbesplitinto16×16,16×8,8×16,or8×8.When8×8modeischosen,each8×8blockcanfurtherbedividedinto8×8,8×4,4×8,or4×4.ThepartitionswithinanMBhavealotofpossiblecombinationsandconsumehugecomputationalpower.Eachblockcanbecompensatedbyusinganindividualmotionvector.InSVC,theexhaustivetechniqueisemployedtoselectthebestmodeforeachMBbutittakeshugeamountofcomputationalcomplexityonthepartoftheencoder.
Inthispaper,afastmodedecisionbasedonprobabilisticanalysisisproposedtospeeduptheSVCencodingprocess.Becausehighcorrelationsexistbetweenthebaselayer(BL)andtheEL,thecodingmodeoftheELcanbepredictedthroughtheBL.To ndouttherelationbetweeneachlayer,severalstatisticsareperformed.First,therelationofmodedistributionbetweentheBLandELareanalyzedtodeter-minethebestmodeforthecurrentMBintheEL.IfthebestmodeforthecurrentMBintheELisdesignatedasmode16×16,themotionvectorinformationintheBLisfurtherusedtodecidewhetherthecurrentmodeshouldbe
c2010IEEE1051-8215/$26.00
Rate-scalable video coding using a zerotree ...A motion compensation algorithm consists of two ...J. Delp, “A tree structured Bayesian scalar ...
Bayesian image restorati... 4页 免费喜欢此文档... Scalable Image and Video Coding Using Wavelet ...indexing algorithm for subband coded video databases...
Bayesian image restorati... 4页 免费喜欢此文档... Scalable Image and Video Coding Using Wavelet ...indexing algorithm for subband coded video databases...
热门文档