Fast Mode Decision Using All-Zero Block Detction for Fidelity and Spatial Scalabale Video Coding
IEEETRANSACTIONSONCIRCUITSANDSYSTEMSFORVIDEOTECHNOLOGY,VOL.20,NO.2,FEBRUARY2010201
FastModeDecisionUsingAll-ZeroBlockDetectionforFidelityandSpatialScalable
VideoCoding
Seung-WonJung,Seung-JinBaek,Chun-SuPark,StudentMember,IEEE,andSung-JeaKo,SeniorMember,IEEE
Abstract—Inscalablevideocoding(SVC)asanextensionofH.264/advancedvideocoding(AVC),acomputationallyexpensiveexhaustivemodedecisionisemployedtoselectthebestcodingmodeforeachmacroblock(MB).Inordertoreducecomputa-tionalcomplexity,weproposeafastmodedecisionalgorithmforSVCwhichusesanall-zeroblock(AZB)detection.Basedontheempiricalanalysisoftheinter-layercorrelationoftheAZB,theMBattheenhancementlayer(EL)ispredictedwhethertobetheAZB.Then,onlypredictedMBsareexaminedbytheAZBdetectionalgorithm.SincethemodedecisioncanbeterminatedbydetectingtheAZB,wedeterminetheprocessingorderofthecandidatemodesattheELinordertoterminatethemodedecisionintheearlystage.Theproposedalgorithmcanbecombinedwithotherconventionalfastmodedecisionmethodsinordertofurtherreducethecomputationalcomplexityofthosemethods.Experimentalresultsshowthattheproposedalgorithmcansigni cantlyspeeduptheencodingprocess,especiallyforlowbit-ratevideosequences,withoutdeterioratingthecodingef ciencyofSVC.
IndexTerms—All-zeroblock(AZB),fastmodedecision,inter-layercorrelation,scalablevideocoding(SVC).
I.Introduction
S
CALABLEvideocoding(SVC)hasbeenactivelyre-searchedandwasrecentlystandardizedbythejointvideoteam(JVT)oftheinternationaltelecommunicationunion-telecommunication(ITU-T)videocodingexpertsgroup(VCEG)andtheInternationalOrganizationforStandardiza-tion(ISO)/InternationalElectrotechnicalCommission(IEC)movingpictureexpertsgroup(MPEG)[1].SVCprovidesascalablebitstreamwhichsupportsmultiplespatial,temporal,and delityscalabilitywithoutproducingasigni cantlossinthecodingef ciency.Thescalabilityisrealizedbymeansofalayeredcodingarchitectureconsistingofthebaselayer(BL)andtheenhancementlayer(EL).Inordertoimprovethecodingef ciencyattheEL,theinter-layercorrelationbetweentheBLandELhasbeenextensivelyutilizedintheliterature[2],[3].
ManuscriptreceivedJuly31,2008;revisedDecember4,2008andMarch17,2009.FirstversionpublishedSeptember1,2009;currentversionpublishedFebruary5,2010.ThisresearchwassupportedbySeoulFutureContentsConvergenceCluster,establishedbySeoulR&BDProgram.ThispaperwasrecommendedbyAssociateEditorZ.He.
TheauthorsarewiththeDepartmentofElectricalEngineering,KoreaUniversity,Seoul136-701,Korea(e-mail:jungsw@dali.korea.ac.kr;sjbaek@dali.korea.ac.kr;cspark@korea.ac.kr;sjko@korea.ac.kr).DigitalObjectIdenti er10.1109/TCSVT.2009.2031387
Thehighcodingef ciencyofSVCisobtainedattheexpenseofthecomputationalcomplexity.FortheBLwhichiscompatibletotheH.264/advancedvideocoding(AVC)standard[4],thecomputationalcomplexityislargelyduetothespatialpredictionintheintracodingandmotionestimation(ME)withvariableblocksizeintheintercoding.WhenencodingtheEL,additionalpredictionmodesareincludedtoutilizetheinter-layercorrelationbetweentheBLandEL[5].Sincetheencoderexaminesallthepossiblemodesandchoosesthebestonewiththeminimumratedistortioncost(RDcost),acomplexityreductiontechniqueforthemodedecisionisdesirablefortheimplementationofSVC.
ItwasreportedthatthemodedistributionattheELishighlycorrelatedwiththatattheBL[6].Therefore,whenencodingthemacroblock(MB)attheEL,thenumberofcandidatemodescanbereducedbyusingthemodeoftheBLwhichisavailableattheencodingoftheEL.In[7],thecandidatemodesareexplicitlydeterminedbyalookuptable.GiventhebestmodeoftheBLandthequantizationparameter(QP),thereducednumberofcandidatemodesisobtainedbyreferringtothelookuptable.Theconventionalfastmodedecisionalgorithms[6],[7]ef cientlyreducethecomputationalcomplexitywithoutcompromisingthecodingef ciencyofSVC.
Thestudyonthediscretecosinetransform(DCT)coef -cientsdemonstratedthattheDCTandquantizationstepscanbeskippedwhenanall-zeroblock(AZB)isexpected[8],[9],[10].Recently,itwasshownin[11]thatanAZBdetectionalgorithmcanbeusedfortheearlyterminationofmotionestimation(EMET)andtheearlyterminationofmodedecision(EMDT).Inthispaper,weapplytheAZBdetectiontechnique[11]tothefastmodedecisionofSVC.SincetheBLandELcontainthesamevideosequencewithdifferentqualityorspatialresolutions,thepositionoftheAZBbetweentheBLandELtendstohaveahighcorrelation.Basedonthisconsideration,weintroduceadecisionrulewhichenablestheAZBdetectionalgorithmtobeadaptivelyappliedtotheEL.Also,basedontheanalysisofthelikelihoodoftheAZBattheEL,thecandidatemodesareprioritizedandsequentiallyexaminedsothattheAZBcanbedetectedintheearliermode.Themainadvantageoftheproposedalgorithmisitscom-patibilitytotheotherfastmodedecisionalgorithms[6],[7].Experimentalresultsdemonstratethattheproposedalgorithm
c2010IEEE1051-8215/$26.00
热门文档