多媒体技术chapter-12-MPEG-Video-Coding-II

2020/2/211chapter12MPEGVideoCodingII12.1OverviewofMPEG-412.2Object-basedVisualCodinginMPEG-412.3SyntheticObjectCodinginMPEG-412.4MPEG-4Objecttypes,ProfileandLevels12.5MPEG-4Part10/H.26412.6MPEG-712.7MPEG-2112.8FurtherExplorationChapter122020/2/21212.1OverviewofMPEG-4MPEG-4:anewerstandard.Besidescompression,paysgreatattentiontoissuesaboutuserinteractivities.MPEG-4departsfromitspredecessorsinadoptinganewobject-basedcoding:Offeringhighercompressionratio,alsobeneficialfordigitalvideocomposition,manipulation,indexing,andretrieval.Figure12.1illustrateshowMPEG-4videoscanbecomposedandmanipulatedbysimpleoperationsonthevisualobjects.Chapter122020/2/213Thebit-rateforMPEG-4videonowcoversalargerangebetween5kbpsto10Mbps.Chapter122020/2/214Chapter122020/2/215OverviewofMPEG-4(Cont'd)MPEG-4(Fig.12.2(b))isanentirelynewstandardfor:(a)Composingmediaobjectstocreatedesirableaudiovisualscenes.(b)MultiplexingandsynchronizingthebitstreamsforthesemediadataentitiessothattheycanbetransmittedwithguaranteedQoS.(c)InteractingwiththeaudiovisualsceneatthereceivingendChapter122020/2/216Chapter122020/2/217OverviewofMPEG-4(Cont'd)ThehierarchicalstructureofMPEG-4visualbitstreamsisverydifferentfromthatofMPEG-1and-2,itisverymuchvideoobject-oriented.Chapter122020/2/21812.2Object-basedVisualCodinginMPEG-4VOP-basedvs.Frame-basedCodingMPEG-1and-2donotsupporttheVOPconcept,andhencetheircodingmethodisreferredtoasframe-based(alsoknownasBlock-basedcoding).Fig.12.4(c)illustratesapossibleexampleinwhichbothpotentialmatchesyieldsmallpredictionerrorsforblock-basedcoding.Fig.12.4(d)showsthateachVOPisofarbitraryshapeandideallywillobtainauniquemotionvectorconsistentwiththeactualobjectmotion.Chapter122020/2/219Fig.12.4:ComparisonbetweenBlock-basedCodingandObject-basedCoding.Chapter122020/2/2110VOP-basedCodingMPEG-4VOP-basedcodingalsoemploystheMotionCompensationtechnique:AnIntra-framecodedVOPiscalledanI-VOP.TheInter-framecodedVOPsarecalledP-VOPsifonlyforwardpredictionisemployed,orB-VOPsifbi-directionalpredictionsareemployed.ThenewdifficultyforVOPs:mayhavearbitraryshapes,shapeinformationmustbecodedinadditiontothetextureoftheVOP.Chapter122020/2/2111VOP-basedMotionCompensation(MC)MC-basedVOPcodinginMPEG-4againinvolvesthreesteps:(a)MotionEstimation.(b)MC-basedPrediction.(c)Codingofthepredictionerror.OnlypixelswithintheVOPofthecurrent(Target)VOPareconsideredformatchinginMC.TofacilitateMC,eachVOPisdividedintomanyMBs.MBsarebydefault16*16inluminanceimagesand8*8inchrominanceimages.Chapter122020/2/2112MPEG-4definesarectangularboundingboxforeachVOP(seeFig.12.5fordetails).TheMBsthatareentirelywithintheVOParereferredtoasInteriorMBs.TheMBsthatstraddle(跨越)theboundaryoftheVOParecalledBoundaryMBs.TohelpmatchingeverypixelinthetargetVOPandmeetthemandatoryrequirementofrectangularblocksintransformcoding(e.g.,DCT),apre-processingstepofpaddingisappliedtotheReferenceVOPspriortomotionestimation.Chapter122020/2/2113Chapter122020/2/2114Chapter122020/2/2115Chapter122020/2/2116II.MotionVectorCodingChapter122020/2/2117Texture（纹理）CodingTexturecodinginMPEG-4canbebasedon:DCTorShapeAdaptiveDCT(SA-DCT).I.TexturecodingbasedonDCTInI-VOP,thegrayvaluesofthepixelsineachMBoftheVOParedirectlycodedusingtheDCTfollowedbyVLC,similartowhatisdoneinJPEG.InP-VOPorB-VOP,MC-basedcodingisemployed--itisthepredictionerrorthatissenttoDCTandVLC.Chapter122020/2/2118CodingfortheInteriorMBs:EachMBis16*16intheluminanceVOPand8*8inthechrominanceVOP.Predictionerrorsfromthesix8*8blocksofeachMBareobtainedaftertheconventionalmotionestimationstep.CodingforBoundaryMBs:ForportionsoftheBoundaryMBsintheTargetVOPoutsideoftheVOP,zerosarepaddedtotheblocksenttoDCTsinceideallypredictionerrorswouldbenearzeroinsidetheVOP.AfterMC,texturepredictionerrorswithintheTargetVOPareobtained.Chapter122020/2/2119II.SA-DCTbasedcodingforBoundaryMBsShapeAdaptiveDCT(SA-DCT)isanothertexturecodingmethodforboundaryMBs.Duetoitseffectiveness,SA-DCThasbeenadoptedforcodingboundaryMBsinMPEG-4Version2.Itusesthe1DDCT-Ntransformanditsinverse,IDCT-N:1DDCT-N:Chapter122020/2/21201DIDCT-N:Chapter122020/2/2121Chapter122020/2/2122ShapeCodingMPEG-4supportstwotypesofshapeinformation,binaryandgrayscale.Binaryshapeinformationcanbeintheformofabinarymap(alsoknownasbinaryalphamap)thatisofthesizeastherectangularboundingboxoftheVOP.Avalue`1'(opaque)or`0'(transparent)inthebitmapindicateswhetherthepixelisinsideoroutsidetheVOP.Chapter122020/2/2123Alternatively,thegray-scaleshapeinformationactuallyreferstothetransparencyoftheshape,withgrayvaluesrangingfrom0(completelytransparent)to255(opaque).Chapter122020/2/2124I.BinaryShapeCodingBABs(BinaryAlphaBlocks):toencodethebinaryalphamapmoreefficiently,themapisdividedinto1616.BlocksItistheboundaryBABsthatcontainthecontourandhencetheshapeinformationfortheVOP|thesubjectofbinaryshapecoding.Twobitmap-basedalgorithms:(a)ModifiedModifiedREAD(MMR).(b)Context-basedArithmeticEncoding(CAE).Chapter122020/2/2125ModifiedModifiedREAD(MMR)MMRisbasicallyaseriesofsimplificationsoftheRelativeElementAddressDesignate(READ)algorithmTheREADalgorithmstart