MPEG-4视频编解码的英文文献

18Int.J.ComputationalVisionandRobotics,Vol.2,No.1,2011Copyright©2011InderscienceEnterprisesLtd.Towardsareal-timevideocompressionbasedonMPEG4AVC/H.264MohamedBoulakradeche*andSamyAit-AoudiaEcoleNationaleSupérieured’Informatique(ESi),BP.68MOuedSmar,16309,ElHarrach,Alger,AlgeriaFax:[phone]E-mail:[e-mail]-mail:[e-mail]*CorrespondingauthorAbstract:MPEG4isanISO/IECstandarddevelopedbymovingpictureexpertsgroup(MPEG).ThestandardvideoMPEG-4visual(part2)introducesthenotionofvideoobjects(VO).Basically,theconceptoftheMPEG-4compressionreliesoncontents.Ontheotherhand,thestandardMPEG-4AVC/H.264(part10)reliesonthesamearchitectureusedinMPEG-2,whichdoesnotintroducetheconceptofvideoobjects.Theperformancesofthesevideocodersdependmainlyonmotionestimation.TheobjectiveofthisworkistoconceiveacodecbasedonstandardMPEG-4AVCintroducingtheconceptofvideoobjectsfromanAVCcodec,whilebeingbasedonmanymotionestimatingtechniques.Thisisdonewiththehopeofreachingthehighestreductionoftheexecutiontimebykeepingthebestpossibleprecisioninordertobeabletointegrateitinremotemonitoringsystem.Keywords:motionestimation;real-timeMPEG-4AVC;object-basedMPEG-4;remotemonitoringsystem.Referencetothispapershouldbemadeasfollows:Boulakradeche,M.andAit-Aoudia,S.(2011)‘Towardsareal-timevideocompressionbasedonMPEG4AVC/H.264’,Int.J.ComputationalVisionandRobotics,Vol.2,No.1,pp.18–33.Biographicalnotes:BoulakradecheMohamedreceivedhisBSinSoftwareComputingandMSinArtificialIntelligenceatAnnabaUniversity,Annaba,Algeriain1984and1994,respectively.HeiscurrentlypursuinghisPhDinInformationEngineeringattheNationalComputerScienceInstituteAlgiers,Algeria.HisresearchinterestsincludepatternrecognitionandimageprocessingSamyAit-AoudiareceivedhisDEA‘Diplômed’EtudesApprofondies’inImageProcessingfromSaint-EtienneUniversity,France,in1990.HereceivedhisPhDinComputerSciencefromtheEcoledesMines,Saint-Etienne,France,in1994.Heiscurrentlya‘MaîtredeConférences’attheNationalComputerScienceInstituteinAlgeria.HeteachesdifferentmodulesatbothBScandMSclevelsincomputerscienceandsoftwareengineering.Hisareasofresearchincludeimageprocessingandconstraintsmanagementinsolidmodelling.Towardsareal-timevideocompressionbasedonMPEG4AVC/H.264191IntroductionOneofthekeyvideocodingparametersisthecoderprocedure,whichiscarriedoutfromthemotionestimatingprocessandthechoiceofthevideocodingmodevideocoding.Theoperatingmodesaregenerallyassociatedwithflow-distortioncharacteristicswhichdependentonthesignal.Theflow-distortionchoicesmadearepartoftheprocessofchoosingthecodingparameters.Thustheoptimisationconsistsinchoosingforeachimagearea,thebestcodedrepresentationconsideringflow-distortion.Thegoalofourworkistooptimisetheservicesofthedataprocessingsequenceofavideosignalwithanaimtoobtainthebestpossiblecompressionratio.2CodecdescriptionThefollowingfiguresgiveageneraldescriptionofthecodec.TheoperationsofanAVCvideocodecaretakenfromvideoobjects(VO)aftermotiondetection.Thevariablecontextadaptivelengthcoding(CAVLC)(RichardsonandZhao,2002)techniqueisappliedforentropiccoding.Figure1Generaldescriptionoftheencoder(seeonlineversionforcolours)20M.BoulakradecheandS.Ait-AoudiaFigure2Generaldescriptionofthedecoder(seeonlineversionforcolours)3CodecconceptionAssumingthatthebackground(Sprite)ofalltheprocessedvideoimagesisthesameandtakingintoaccountthehypothesisoftheinvarianceofthegreylevels,wewillseealltheactions,inasimplifiedform,oftheobject-basedAVCcodec.3.1SamplingandcuttingBeforepresentinganyimagetothecodec,aconversiontowardsspaceY(brightness)U(bluechrominance)andV(redchrominance)isappliedinordertoreducetohalfthesizeoftheimagetobecoded.Thesampling4:2:0isselected,withaninitialrepresentationoftheimagepermacro-blocksof16×16pixelsandacuttinginblocksof8×8formotionestimating.Blocksofsize4×4areselectedinthediscretecosinetransformation(DCT)transformfortheirgoodcorrelationinter-pixelandtheirlowcostcomputingtime.ThepicturesarenaturallyrepresentedinRGBformat(everypixelhasthreecomponentsred,greenandblue).ThefollowingformulasareusedtoswitchtowardstheYUVspace(MNSD2008).()()()298*(16)409*(128)1288298*(16)100*(128)208*(128)1288298*(16)516*(128)1288;=−+−+=−−−−−+=−+−+RYVGYUVBYU(1)Towardsareal-timevideocompressionbasedonMPEG4AVC/H.26421()()()()()()66*129*25*1288638*74*112*1288128112*94*18*1288128=++++=−−+++=−−++YRGBURGBVRGB(2)Inordertoreducethesizeoftheimagetobecodedbyhalfthesampling4:2:0istobeapplied.Itsprincipleistosampleonebluechrominanceandoneredchrominanceeveryfour-greenchrominance.3.2ObjectdetectionThestepofmotionsegmentationconsistsoftheseparationofeachimageinseveralparts,thusformingthesegmentationmapwhichgivesthespacepositionofeachobjectduringthesequencetobetreatedbyusingtheframedifferencealgorithm(Brisbarreetal.,2003).Thisalgorithmusesdifferentsimilaritycriteriasuchasthesumofabsolutedifferences(SAD).Letusnotethatthenumberandthesizeoftheobjectsinasequencearedynamic.3.3ObjectstructureAvideosequence(inourcodec)ismadeupofSprite,aseriesofobjects(VOP)withtheiroutlines.Eachobjecthasanumber(VOC)whichisusedwhilereorderingt