多媒体技术chapter-10-Basic-Video-Compression-Techniques

2020/2/211Chapter10BasicVideoCompressionTechniquesChapter102020/2/21210.1IntroductiontoVideoCompression10.2VideoCompressionwithMotionCompensation10.3SearchforMotionVectors10.4H.26110.5H.263Chapter102020/2/21310.1IntroductiontoVideoCompressionAvideoconsistsofatime-orderedsequenceofframes,i.e.,images.Anobvioussolutiontovideocompressionwouldbepredictivecodingbasedonpreviousframes.Compressionproceedsbysubtractingimages:subtractintimeorderandcodetheresidualerror.Itcanbedoneevenbetterbysearchingforjusttherightpartsoftheimagetosubtractfromthepreviousframe.Chapter102020/2/21410.2VideoCompressionwithMotionCompensationConsecutiveframesinavideoaresimilar----temporalredundancyexists.Temporalredundancyisexploitedsothatnoteveryframeofthevideoneedstobecodedindependentlyasanewimage.Thedifferencebetweenthecurrentframeandotherframe(s)inthesequencewillbecoded----smallvaluesandlowentropy,goodforcompression.Chapter102020/2/215StepsofVideocompressionbasedonMotionCompensation(MC):1.MotionEstimation(motionvectorsearch).2.MC-basedPrediction.3.Derivationofthepredictionerror,i.e.,thedifference.Chapter102020/2/216MotionCompensationEachimageisdividedintomacroblocksofsizeN×N.Motioncompensationisperformedatthemacroblocklevel.--ThecurrentimageframeisreferredtoasTargetFrame.--AmatchissoughtbetweenthemacroblockintheTargetFrameandthemostsimilarmacroblockinpreviousand/orfutureframe(s)(referredtoasReferenceframe(s)).Chapter102020/2/217MotionCompensation--ThedisplacementofthereferencemacroblocktothetargetmacroblockiscalledamotionvectorMV.--MVsearchisusuallylimitedtoasmallimmediateneighborhoodBothhorizontalandverticaldisplacementsintherange[-p,p].Thismakesasearchwindowofsize(2p+1)×(2p+1).Chapter102020/2/218Chapter102020/2/21910.3SearchforMotionVectorsThedifferencebetweentwomacroblockscanthenbemeasuredbytheirMeanAbsoluteDifference(MAD):Chapter102020/2/2110Thegoalofthesearchistofindavector(i,j)asthemotionvectorMV=(u,v),suchthatMAD(i,j)isminimum:Chapter102020/2/211110.3.1SequentialSearchsequentiallysearchthewhole(2p+1)×(2p+1)windowintheReferenceframe--AMBcenteredateachofthepositionswithinthewindowiscomparedtotheMBintheTargetframepixelbypixelandtheirrespectiveMADisthenderivedusingEq.(10.1).--Thevector(i,j)thatofferstheleastMADisdesignatedastheMV(u,v)fortheMBintheTargetframe.Chapter102020/2/2112Chapter102020/2/2113sequentialsearchmethodisverycostlyassumingeachpixelcomparisonrequiresthreeoperations(subtraction,absolutevalue,addition),thecostforobtainingamotionvectorforasinglemacroblockisChapter102020/2/211410.3.22DLogarithmicSearchTheproceduretakesseveraliterationsandisakintoabinarysearch:--InitiallyonlyninelocationsinthesearchwindowareusedasseedsforaMAD-basedsearch;theyaremarkedas`1'.--AftertheonethatyieldstheminimumMADislocated,thecenterofthenewsearchregionismovedtoitandthestep-size(offset)isreducedtohalf.--Inthenextiteration,theninenewlocationsaremarkedas`2',andsoon.Chapter102020/2/2115Chapter102020/2/2116Chapter102020/2/2117Usingthesameexampleasintheprevioussubsection,thetotaloperationspersecondisdroppedto:Chapter102020/2/211810.3.3HierarchicalSearchThesearchcanbenefitfromahierarchical(multiresolution)approachinwhichinitialestimationofthemotionvectorcanbeobtainedfromimageswithasignificantlyreducedresolution.Figure10.3:athree-levelhierarchicalsearchinwhichtheoriginalimageisatLevel0,imagesatLevels1and2areobtainedbydown-samplingfromthepreviouslevelsbyafactorof2,andtheinitialsearchisconductedatLevel2.Chapter102020/2/2119Chapter102020/2/2120Chapter102020/2/2121Chapter102020/2/2122Chapter102020/2/212310.4H.261Anearlierdigitalvideocompressionstandard,itsprincipleofMC-basedcompressionisretainedinalllatervideocompressionstandards.--Thestandardwasdesignedforvideophone,videoconferencingandotheraudiovisualservicesoverISDN.--Thevideocodecsupportsbit-ratesofp*64kbps,whereprangesfrom1to30(Hencealsoknownasp*64).--Requirethatthedelayofthevideoencoderbelessthan150msecChapter102020/2/2124ITURecommendations&H.261VideoFormats1.H.221--Framestructureforanaudiovisualchannelsupporting64to1,920kbps.2.H.230--Framecontrolsignalsforaudiovisualsystems.3.H.242--Audiovisualcommunicationprotocols.4.H.261--Videoencoder/decoderforaudiovisualservicesatp*64kbps.5.H.320--Narrow-bandaudiovisualterminalequipmentforp*64kbpstransmission.Chapter102020/2/2125Chapter102020/2/2126Chapter102020/2/2127H.261FrameSequenceTwotypesofimageframesaredefined:Intra-frames(I-frames)andInter-frames(P-frames):--I-framesaretreatedasindependentimages.--P-framesarenotindependent:codedbyaforwardpredictivecodingmethod--TemporalredundancyremovalisincludedinP-framecoding--Toavoidpropagationofcodingerrors,anI-frameisusuallysentacoupleoftimesineachsecondofthevideo.Chapter102020/2/2128MotionvectorsinH.261arealwaysmeasuredinunitsoffullpixelandtheyhavealimitedrangeof±15pixelsi.e.,p=15.Chapter102020/2/212910.4.1Intra-frame(I-frame)CodingChapter102020/2/2130Macroblocksareofsize16*16pixelsfortheYframe,and8*8forCbandCrframes,since4:2:0chromasubsamplingisemployed.AmacroblockconsistsoffourY,oneCb,andoneCr8*8blocks.Fore