英文翻译.docx
《英文翻译.docx》由会员分享,可在线阅读,更多相关《英文翻译.docx(24页珍藏版)》请在冰豆网上搜索。
英文翻译
郑州轻工业学院
本科毕业设计(论文)
英文翻译
题目MPEG-2视频压缩
学生姓名
专业班级电子信息科学与技术2004级1班
学号
院(系)计算机与通信工程学院
指导教师(职称)
完成时间2008年4月15日
英文原文
MPEG-2VIDEOCOMPRESSION
byP.N.Tudor.
ThispaperwontheIEEJLanghamThompsonPrize.
Electronics&CommunicationEngineeringJournal,December1995.
MPEG-2isanextensionoftheMPEG-1internationalstandardfordigitalcompressionofaudioandvideosignals.MPEG-1wasdesignedtocodeprogressivelyscannedvideoatbitratesuptoabout1.5Mbit/sforapplicationssuchasCD(compactdiscinteractive).MPEG-2isdirectedatbroadcastformatsathigherdatarates;itprovidesextraalgorithmic'tools'forefficientlycodinginterlacedvideo,supportsawiderangeofbitratesandprovidesformultichannelsurroundsoundcoding.ThistutorialpaperintroducestheprinciplesusedforcompressingvideoaccordingtotheMPEG-2standard,outlinesthegeneralstructureofavideocoderanddecoder,anddescribesthesubsets('profiles')ofthetoolkitandthesetsofconstraintsonparametervalues('levels')definedtodate.
1VIDEOFUNDAMENTALS
TelevisionservicesinEuropecurrentlybroadcastvideoataframerateof25Hz.Eachframeconsistsoftwointerlacedfields,givingafieldrateof50Hz.Thefirstfieldofeachframecontainsonlytheoddnumberedlinesoftheframe(numberingthetopframelineasline1).Thesecondfieldcontainsonlytheevennumberedlinesoftheframeandissampledinthevideocamera20msafterthefirstfield.Itisimportanttonotethatoneinterlacedframecontainsfieldsfromtwoinstantsintime.Americantelevisionissimilarlyinterlacedbutwithaframerateofjustunder30Hz.
Invideosystemsotherthantelevision,non-interlacedvideoiscommonplace(forexample,mostcomputersoutputnon-interlacedvideo).Innon-interlacedvideo,allthelinesofaframearesampledatthesameinstantintime.Non-interlacedvideoisalsotermed'progressivelyscanned'or'sequentiallyscanned'video.
Thered,greenandblue(RGB)signalscomingfromacolourtelevisioncameracanbeequivalentlyexpressedasluminance(Y)andchrominance(UV)components.Thechrominancebandwidthmaybereducedrelativetotheluminancewithoutsignificantlyaffectingthepicturequality.Forstandarddefinitionvideo,CCIRrecommendationdefineshowthecomponent(YUV)videosignalscanbesampledanddigitisedtoformdiscretepixels.Theterms4:
2:
2and4:
2:
0areoftenusedtodescribethesamplingstructureofthedigitalpicture.4:
2:
2meansthechrominanceishorizontallysubsampledbyafactoroftworelativetotheluminance;4:
2:
0meansthechrominanceishorizontallyandverticallysubsampledbyafactoroftworelativetotheluminance.
MPEG-2iscapableofcompressingthebitrateofstandard-definition4:
2:
0videodowntoabout3-15Mbit/s.Atthelowerbitratesinthisrange,theimpairmentsintroducedbytheMPEG-2codinganddecodingprocessbecomeincreasinglyobjectionable.Fordigitalterrestrialtelevisionbroadcastingofstandard-definitionvideo,abitrateofaround6Mbit/sisthoughttobeagoodcompromisebetweenpicturequalityandtransmissionbandwidthefficiency.
2BITRATEREDUCTIONPRINCIPLES
Abitratereductionsystemoperatesbyremovingredundantinformationfromthesignalatthecoderpriortotransmissionandre-insertingitatthedecoder.Acoderanddecoderpairarereferredtoasa'codec'.Invideosignals,twodistinctkindsofredundancycanbeidentified.
Spatialandtemporalredundancy:
Pixelvaluesarenotindependent,butarecorrelatedwiththeirneighboursbothwithinthesameframeandacrossframes.So,tosomeextent,thevalueofapixelispredictablegiventhevaluesofneighbouringpixels.
Psychovisualredundancy:
Thehumaneyehasalimitedresponsetofinespatialdetail,andislesssensitivetodetailnearobjectedgesoraroundshot-changes.Consequently,controlledimpairmentsintroducedintothedecodedpicturebythebitratereductionprocessshouldnotbevisibletoahumanobserver.
TwokeytechniquesemployedinanMPEGcodecareintra-frameDiscreteCosineTransform(DCT)codingandmotion-compensatedinter-frameprediction.ThesetechniqueshavebeensuccessfullyappliedtovideobitratereductionpriortoMPEG,notablyfor625-linevideocontributionstandardsat34Mbit/sandvideoconferencesystemsatbitratesbelow2Mbit/s.
Intra-frameDCTcoding
DCT:
Atwo-dimensionalDCTisperformedonsmallblocks(8pixelsby8lines)ofeachcomponentofthepicturetoproduceblocksofDCTcoefficients(Fig.1).ThemagnitudeofeachDCTcoefficientindicatesthecontributionofaparticularcombinationofhorizontalandverticalspatialfrequenciestotheoriginalpictureblock.ThecoefficientcorrespondingtozerohorizontalandverticalfrequencyiscalledtheDCcoefficient.
Fig.1Thediscretecosinetransform(DCT).
PixelvalueandDCTcoefficientmagnitudearerepresentedbydotsize.
TheDCTdoesn'tdirectlyreducethenumberofbitsrequiredtorepresenttheblock.Infactforan8x8blockof8bitpixels,theDCTproducesan8x8blockof11bitcoefficients(therangeofcoefficientvaluesislargerthantherangeofpixelvalues.)Thereductioninthenumberofbitsfollowsfromtheobservationthat,fortypicalblocksfromnaturalimages,thedistributionofcoefficientsisnon-uniform.Thetransformtendstoconcentratetheenergyintothelow-frequencycoefficientsandmanyoftheothercoefficientsarenear-zero.Thebitratereductionisachievedbynottransmittingthenear-zerocoefficientsandbyquantisingandcodingtheremainingcoefficientsasdescribedbelow.Thenon-uniformcoefficientdistributionisaresultofthespatialredundancypresentintheoriginalimageblock.
Quantisation:
ThefunctionofthecoderistotransmittheDCTblocktothedecoder,inabitrateefficientmanner,sothatitcanperformtheinversetransformtoreconstructtheimage.IthasbeenobservedthatthenumericalprecisionoftheDCTcoefficientsmaybereducedwhilestillmaintaininggoodimagequalityatthedecoder.Quantisationisusedtoreducethenumberofpossiblevaluestobetransmitted,reducingtherequirednumberofbits.
Coding:
TheserialisationandcodingofthequantisedDCTcoefficientsexploitsthelikelyclusteringofenergyintothelow-frequencycoefficientsandthefrequentoccurrenceofzero-valuecoefficients.TheblockisscannedinadiagonalzigzagpatternstartingattheDCcoefficienttoproducealistofquantisedcoefficientvalues,orderedaccordingtothescanpattern.
Thelistofvaluesproducedbyscanningisentropycodedusingavariable-lengthcode(VLC).EachVLCcodeworddenotesarunofzerosfollowedbyanon-zerocoefficientofaparticularlevel.VLCcodingrecognisesthatshortrunsofzerosaremorelikelythanlongonesandsmallcoefficientsaremorelikelythanlargeones.TheVLCallocatescodewordswhichhavedifferentlengthsdependingupontheprobabilitywithwhichtheyareexpectedtooccur.Toenablethedecodertodistinguishwhereonecodeendsandthenextbegins,theVLChasthepropertythatnocompletecodeisaprefixofanyother.
Fig.1showsthezigzagscanningprocess,usingthescanpatterncommontobothMPEG-1andMPEG-2.MPEG-2hasanadditional'alternate'scanpatternintendedforscanningthequantisedcoefficientsresultingfrominterlacedsourcepictures.
Toillustratethevariable-lengthcodingprocess,considerthefollowingexamplelistofvaluesproducedbyscanningthequantisedcoefficientsfromatransformedblock:
12,6,6,0,4,3,0,0,0...0
Thefirststepistogroupthevaluesintorunsof(zeroormore)zerosfollowedbyanon-zerovalue.Additionally,thefinalrunofzerosisreplacedwithanendofblock(EOB)marker.Usingparenthesestoshowthegroups,thisgives:
(12),(6),(6),(0,4),(3)EOB
Thesecondstepistogeneratethevariablelengthcodewordscorrespondingtoeachgroup(arunofzerosfollowedbyanon-zerovalue)andtheEOBmarker.Table1showsanextractoftheDCTcoefficientVLCtablecommontobothMPEG-1andMPEG-2.MPEG-2hasanadditional'intra'VLCoptimisedforcodingintrablocks(seeSection4).UsingthevariablelengthcodefromTable1andaddingspacesandcommasforreadability,thefinalcodedrepresentationoftheexampleblockis:
00000000110100,001000010,001000010,00000011000,001010,10
Table1:
ExtractfromtheMPEG-2DCTcoefficientVLCtable.
Lengthofrunofzeros
Valueofnon-zero
coefficient
Variable-lengthcodeword
0
12
00000000110100
0
6
001000010
1
4
00000011000
0
3
001010
EOB
-
10
Motion-compensatedinter-frameprediction
Thistechniqueexploitstemporalredundancybyattemptingtopredicttheframetobecodedfromaprevious'reference'frame.Thepredictioncannotbebasedonasourcepicturebecausethepredictionhastoberepeatableinthedecoder,wherethesourcepicturesarenotavailable(thedecodedpicturesarenotidenticaltothesourcepicturesbecausethebitratereductionprocessintroducessmalldistortionsintothedecodedpicture.)Consequently,thecodercontainsalocaldecoderwhichreconstructspicturesexactlyastheywouldbeinthedecoder,fromwhichpredictionscanbeformed.
3MPEG-2DETAILS
Codecstructure
InanMPEG-2system,theDCTandmotion-compensatedinterframepredictionarecombined,asshowninFig.2.Thecodersubtractsthemotion-compensatedpredictionfromthesourcepicturetoforma'predictionerror'picture.ThepredictionerroristransformedwiththeDCT,thecoefficientsarequantisedandthesequantisedvaluescodedusingaVLC.Thecodedluminanceandchrominancepredictionerroriscombinedwith'sideinformation'requiredbythede