多媒体技术chapter-11-MPEG-Video-Coding-I--MPEG-1-and-2

2020/2/181Chapter11MPEGVideoCodingI:MPEG-1andMPEG-211.1Overview11.2MPEG-111.3MPEG-2Chapter112020/2/18211.1OverviewMPEG:MovingPicturesExpertsGroup,establishedin1988forthedevelopmentofdigitalvideo.Itisappropriatelyrecognizedthatproprietary（所有权）interestsneedtobemaintainedwithinthefamilyofMPEGstandards:Accomplishedbydefiningonlyacompressedbitstreamthatimplicitlydefinesthedecoder.Thecompressionalgorithms,andthustheencoders,arecompletelyuptothemanufacturers.Chapter112020/2/18311.2MPEG-1MPEG-1adoptstheCCIR601digitalTVformatalsoknownasSIF(SourceInputFormat).MPEG-1supportsonlynon-interlacedvideo.Normally,itspictureresolutionis:352*240forNTSCvideoat30fps352*288forPALvideoat25fpsItuses4:2:0chromasubsamplingChapter112020/2/18411.2.1MotionCompensationinMPEG-1MotionCompensation(MC)basedvideoencodinginH.261worksasfollows:InMotionEstimation(ME),eachMBoftheTargetP-frameisassignedabestmatchingMBfromthepreviouslycodedIorPframePredictionerror:ThedifferencebetweentheMBanditsmatchingMB,senttoDCTanditssubsequentencodingsteps.Thepredictionisfromapreviousframe---forwardprediction.Chapter112020/2/18511.2.1MotionCompensationinMPEG-1Chapter112020/2/186MPEGintroducesaB-framesbi-directionalmotioncompensation.TheMC-basedB-framecodingidea--EachMBfromaB-framewillhaveuptotwomotionvectors(MVs)--Ifmatchinginbothdirectionsissuccessful,thentwoMVswillbesentandthetwocorrespondingmatchingMBsareaveraged--Ifanacceptablematchcanbefoundinonlyoneofthereferenceframes,thenonlyoneMVanditscorrespondingMBwillbeusedfromeithertheforwardorbackwardprediction.Chapter112020/2/187Chapter112020/2/188Chapter112020/2/18911.2.2OtherMajorDifferencesfromH.261Sourceformatssupported:H.261onlysupportsCIF(352*288)andQCIF(176*144)sourceformats,MPEG-1supportsSIF(352*240forNTSC,352*288forPAL).MPEG-1alsoallowsspecificationofotherformatsaslongastheConstrainedParameterSet(CPS)asshowninTable11.1issatisfiedChapter112020/2/1810Chapter112020/2/1811InsteadofGOBsasinH.261,anMPEG-1picturecanbedividedintooneormoreslicesMaycontainvariablenumbersofmacroblocksinasinglepicture.Mayalsostartandendanywhereaslongastheyfillthewholepicture.Eachsliceiscodedindependently---additionalflexibilityinbit-ratecontrol.Sliceconceptisimportantforerrorrecovery.OtherMajorDifferencesfromH.261(Cont'd)Chapter112020/2/1812Chapter112020/2/1813OtherMajorDifferencesfromH.261(Cont'd)Quantization:MPEG-1quantizationusesdifferentquantizationtablesforitsIntraandIntercoding(Table11.2and11.3).ForDCTcoefficientsinIntramode:ForDCTcoefficientsinIntermode,Chapter112020/2/1814Chapter112020/2/1815MPEG-1allowsmotionvectorstobeofsub-pixelprecision(1/2pixel).ThetechniqueofbilinearinterpolationforH.263canbeusedtogeneratetheneededvaluesathalf-pixellocations.Comparedtothemaximumrangeof15pixelsformotionvectorsinH.261,MPEG-1supportsarangeof[-512,511.5]forhalf-pixelprecisionand[-1024,1023]forfull-pixelprecisionmotionvectors.TheMPEG-1bitstreamallowsrandomaccess--accomplishedbyGOPlayerinwhicheachGOPistimecoded.Chapter112020/2/1816TypicalSizesofMPEG-1FramesChapter112020/2/1817Chapter112020/2/18189.1.3AGlanceattheJPEGBitstreamChapter112020/2/1819Chapter112020/2/182011.3MPEG-2MPEG-2:Forhigherqualityvideoatabit-rateofmorethan4Mbps.Definedsevenprofilesaimedatdifferentapplications:Simple,Main,SNRscalable,Spatiallyscalable,High,4:2:2,Multiview.Withineachprofile,uptofourlevelsaredefinedTheDVDvideospecificationallowsonlyfourdisplayresolutions:720*480,704*480,352*480,and352*240Chapter112020/2/1821Chapter112020/2/182211.3.1SupportingInterlacedVideoMPEG-2mustsupportinterlacedvideoaswellsincethisisoneoftheoptionsfordigitalbroadcastTVandHDTV.Ininterlacedvideo,eachframeconsistsoftwofields,referredtoasthetop-fieldandthebottom-field.InaFrame-picture,allscanlinesfrombothfieldsareinterleavedtoformasingleframe,thendividedinto16*16MBsandcodedusingMC.Ifeachfieldistreatedasaseparatepicture,thenitiscalledField-picture.Chapter112020/2/1823Chapter112020/2/1824FiveModesofPredictionsMPEG-2definesFramePredictionandFieldPredictionaswellasfivepredictionmodes:1.FramePredictionforFrame-pictures:IdenticaltoMPEG-1MC-basedpredictionmethodsinbothP-framesandB-frames.2.FieldPredictionforField-pictures:AMBsizeof16*16fromField-picturesisused.Fordetails,seeFig.11.6(b).Chapter112020/2/18253.FieldPredictionforFrame-pictures:Thetop-fieldandbottom-fieldofaFrame-picturearetreatedseparately.Each16*16MBfromthetargetFrame-pictureissplitintotwo16*8parts,eachcomingfromonefield.Fieldpredictioniscarriedoutforthese16*8partsinamannersimilartothatshowninFig.11.6(b).4.16*8MCforField-pictures:Each16*16MBfromthetargetField-pictureissplitintotopandbottom16*8halves.Fieldpredictionisperformedoneachhalf.Thisgeneratestwomotionvectorsforeach16*16MBintheP-Field-picture,anduptofourmotionvectorsforeachMBintheB-Field-picture.Chapter112020/2/18265.Dual-PrimeforP-pictures:First,Fieldpredictionfromeachpreviousfieldwiththesameparity(toporbottom)ismade.Eachmotionvectormvisthenusedtoderiveacalculatedmotionvectorcvinthefieldwiththeoppositeparitytakingintoaccountthetemporalscalingandverticalshi