HEVC核心编码技术

HEVC核心编码技术讲解（英文讲解有翻译）HEVC核心编码技术之一.图像的编码划分IV.HEVC的编码技术HEVCVideoCodingTechniquesAsinallpriorITU-TandISO/IECJTC1videocodingstandardssinceH.261[2],theHEVCdesignfollowstheclassicblock-basedhybridvideocodingapproach(asdepictedinFig.1).Thebasicsource-codingalgorithmisahybridofinterpicturepredictiontoexploittemporalstatisticaldependences,intrapicturepredictiontoexploitspatialstatisticaldependences,andtransformcodingofthepredictionresidualsignalstofurtherexploitspatialstatisticaldependences.ThereisnosinglecodingelementintheHEVCdesignthatprovidesthemajorityofitssignificantimprovementincompressionefficiencyinrelationtopriorvideocodingstandards.Itis,rather,apluralityofsmallerimprovementsthatadduptothesignificantgain.和从H.261以来的视频编码标准一样，HEVC的设计沿用了经典的基于块的混合视频编码方式(如图1所示)。基本的信源编码算法是对时域统计相关性使用帧间预测，对空域统计相关性使用帧内预测，再对预测残差信号使得变换编码进一步去除空间统计相关性。HEVC取得比以前的标准更好的压缩效果，而且以很小的改善就取得了很大的收获。Fig.1.TypicalHEVCvideoencoder(withdecodermodelingelementsshadedinlightgray).A.图像的像素表示SampledRepresentationofPicturesForrepresentingcolorvideosignals,HEVCtypicallyusesatristimulusYCbCrcolorspacewith4:2:0sampling(althoughextensiontoothersamplingformatsisstraightforward,andisplannedtobedefinedinasubsequentversion).ThisseparatesacolorrepresentationintothreecomponentscalledY,Cb,andCr.TheYcomponentisalsocalledluma,andrepresentsbrightness.ThetwochromacomponentsCbandCrrepresenttheextenttowhichthecolordeviatesfromgraytowardblueandred,respectively.Becausethehumanvisualsystemismoresensitivetolumathanchroma,the4:2:0samplingstructureistypicallyused,inwhicheachchromacomponenthasonefourthofthenumberofsamplesofthelumacomponent(halfthenumberofsamplesinboththehorizontalandverticaldimensions).Eachsampleforeachcomponentistypicallyrepresentedwith8or10bofprecision,andthe8-bcaseisthemoretypicalone.Intheremainderofthispaper,wefocusourattentiononthetypicaluse:YCbCrcomponentswith4:2:0samplingand8bpersamplefortherepresentationoftheencodedinputanddecodedoutputvideosignal.为了表示色彩视频信号，HEVC使用了典型的三色色彩空间YCbCr,以及4:2:0像素样本。Y表示亮度，Cb,Cr为色度，分别表示绿，红。每个像素是8比特或10比特表示;ThevideopicturesaretypicallyprogressivelysampledwithrectangularpicturesizesW×H,whereWisthewidthandHistheheightofthepictureintermsoflumasamples.Eachchromacomponentarray,with4:2:0sampling,isthenW/2×H/2.Givensuchavideosignal,theHEVCsyntaxpartitionsthepicturesfurtherasdescribedfollows.视频图像通常是逐行扫描的，图像尺寸为WxH,W和H分别为高度像素的宽度和高度；在4:2:0像素样本中，每个色度分量的宽度和高度分别为W/2和H/2。B.将图像分割成编码树单元(CTU)DivisionofthePictureintoCodingTreeUnitsApictureispartitionedintocodingtreeunits(CTUs),whicheachcontainlumaCTBsandchromaCTBs.AlumaCTBcoversarectangularpictureareaofL×LsamplesofthelumacomponentandthecorrespondingchromaCTBscovereachL/2×L/2samplesofeachofthetwochromacomponents.ThevalueofLmaybeequalto16,32,or64asdeterminedbyanencodedsyntaxelementspecifiedintheSPS.Comparedwiththetraditionalmacroblockusingafixedarraysizeof16×16lumasamples,asusedbyallpreviousITU-TandISO/IECJTC1videocodingstandardssinceH.261(thatwasstandardizedin1990),HEVCsupportsvariable-sizeCTBsselectedaccordingtoneedsofencodersintermsofmemoryandcomputationalrequirements.ThesupportoflargerCTBsthaninpreviousstandardsisparticularlybeneficialwhenencodinghigh-resolutionvideocontent.ThelumaCTBandthetwochromaCTBstogetherwiththeassociatedsyntaxformaCTU.TheCTUisthebasicprocessingunitusedinthestandardtospecifythedecodingprocess.图像被分割成编码树单元(CTU),每个CTU包含亮度CTB和色度CTB;每个高度CTB表示图像中的LxL亮度像素块区域，对应的两个色度CTB的大小为L/2xL/2;L的大小可以是16，32，64，并在SPS的语法元素中声明;相对于传统的16x16大小的宏块，HEVC对可变尺寸的CTB选择需要依据编码端的内存和计算资源来确定;大尺寸的CTB支持，相对于之前的标准，在编码高分辨率视频内容时非常有效果;CTU由一个亮度CTB，两个色度CTB以及相关语法组成;CTU是解码处理的基本处理单元;C.将CTB分割成CBDivisionoftheCTBintoCBsTheblocksspecifiedaslumaandchromaCTBscanbedirectlyusedasCBsorcanbefurtherpartitionedintomultipleCBs.Partitioningisachievedusingtreestructures.ThetreepartitioninginHEVCisgenerallyappliedsimultaneouslytobothlumaandchroma,althoughexceptionsapplywhencertainminimumsizesarereachedforchroma.亮度CTB和色度CTB可以直接当作CB使用;也可以更一步划分成多个CB;划分是以树结构的方式实现;HEVC的树划分方式通常是同时用于亮度和色度;TheCTUcontainsaquadtreesyntaxthatallowsforsplittingtheCBstoaselectedappropriatesizebasedonthesignalcharacteristicsoftheregionthatiscoveredbytheCTB.ThequadtreesplittingprocesscanbeiterateduntilthesizeforalumaCBreachesaminimumallowedlumaCBsizethatisselectedbytheencoderusingsyntaxintheSPSandisalways8×8orlarger(inunitsoflumasamples).CTU包含一个四叉树的语法，它允许对CTB区域依据信号特征划分成合适尺寸的CB;四叉树的切分处理可以迭代进行，直到亮度CB到了允许的最小尺寸，通常是8x8或更大;这个语法元素在SPS中定义;TheboundariesofthepicturearedefinedinunitsoftheminimumallowedlumaCBsize.Asaresult,attherightandbottomedgesofthepicture,someCTUsmaycoverregionsthatarepartlyoutsidetheboundariesofthepicture.Thisconditionisdetectedbythedecoder,andtheCTUquadtreeisimplicitlysplitasnecessarytoreducetheCBsizetothepointwheretheentireCBwillfitintothepicture.图像边界定义为最小可允许的亮度CB大小的整数倍;因此，图像右下角的边缘，有些CTU可能会超出图像的边界;这种情况需要解码器检查并处理，并且CTU的四叉树暗示对CB尺寸的划分最好是整个CB正好适合图像;Fig.3.ModesforsplittingaCBintoPBs,subjecttocertainsizeconstraints.Forintrapicture-predictedCBs,onlyM×MandM/2×M/2aresupported.D.预测块(PB)和预测单元(PU)PBsandPUsThepredictionmodefortheCUissignaledasbeingintraorinter,accordingtowhethe