Color motion artifact detection and processing app.docx

Color motion artifact detection and processing app.docx

《Color motion artifact detection and processing app.docx》由会员分享，可在线阅读，更多相关《Color motion artifact detection and processing app.docx（16页珍藏版）》请在冰豆网上搜索。

Colormotionartifactdetectionandprocessingapp

Colormotionartifactdetectionandprocessingapparatuscompatiblewithvideocodingstandards

BACKGROUNDOFTHEINVENTION

Thepresentinventionrelatestoprocessingofvideodata,andmorespecificallytodetectionandreductionofcolor-motionartifactsinprocessingofvideodata.

Successiveframesinatypicalvideosequenceareoftenverysimilartoeachother.Forexample,asequenceofframesmayhavescenesinwhichanobjectmovesacrossastationarybackground,orabackgroundmovesbehindastationaryobject.Consequently,manyscenesinoneframemayalsoappearinadifferentpositionofasubsequentframe.Videosystemstakeadvantageofsuchredundancywithintheframesbyusingpredictivecodingtechniques,suchasmotionestimationandmotioncompensation,toreducethevolumeofdatarequiredincompressingtheframes.

Inaccordancewiththewell-knownmotionestimationtechnique,toconservebitrate,datarelatedtothedifferencesbetweenpositionsofsimilarobjectsinsuccessiveframesarecapturedbyoneormoremotionvectors.Themotionvectorsarethenusedtoidentifythespatialcoordinatesoftheshiftedobjectsinasubsequentframe.Themotionvectorsthereforelimitthebitratethatwouldotherwiseberequiredtoencodethedataassociatedwiththeshiftedobjects.

MotionestimationandcompensationareusedinseveralinternationalstandardssuchasH.261,H.263,MPEG-1,MPEG-2,andMPEG-4.Partlyduetoitscomputationalintensity,amotionvectorissharedtypicallybyallcolorcomponentsin（Y,U,V）or（Y,Cr,Cb）coordinatesystems.Inthe（Y,U,V）colorcoordinatesystem,Yisthelumacomponent（alsoreferredtobelowastheluminanceandisrelatedtotheintensity）,andUandVarethechromacomponents（alsoreferredtobelowasthechrominancecomponentsandarerelatedtohueandsaturation）ofacolor.Similarly,inthe（Y,Cr,Cb）colorcoordinatesystem,Yisthelumacomponent,andCbandCrarethechromacomponents.Eachmotionvectoristypicallygeneratedforamacroblock.Eachmacroblocktypicallyincludesablockof,e.g.,16×16or8×8pixels.TheMPEG-2standardprovidesaninterlacedmodethatseparateseach16×16macroblockintotwo16×8sub-macroblockseachhavinganassociatedmotionvector.Inthefollowing,thetermsblock,macroblock,andsub-macroblockmaybeusedinterchangeably.

Tosimplifycomputation,mostcommonlyknownvideostandardsuseonlytheluminancecomponenttogenerateamotionvectorforeachmacroblock.Thismotionvectorissubsequentlyappliedtobothchromacomponentsassociatedwiththatmacroblock.Thegenerationofamotionvectorusingonlytheluminancecomponentmaycauseundesirablecolor-motionartifacts（alternativelyreferredtohereinbelowascolorartifacts）suchascolorpatches.

Tofurtherthereducebitrateforencodingofvideodata,inter-frameandintra-frameencodinghavebeendeveloped.Inaccordancewiththeinter-framecoding,thedifferencebetweenthedatacontainedinapreviousframeandacurrentframeisusedtoencodethecurrentframe.Theinter-frameencodingmaynotimprovecodingefficiency,forexample,ifthecurrentframeisthefirstframeinanewscene（i.e.,whenthereisascenechange）,inwhichcaseintra-frameencodingisused.Inaccordancewiththeintra-frameencoding,onlytheinformationcontainedwithintheframeitselfisusedtoencodetheframe.

FIG.1isasimplifiedhigh-levelblockdiagramofaconventionalsystem100adaptedtodetectcolor-motionartifacts.System100receivesasequenceofincomingvideoframesviaframereorderblock102.Inresponse,framereorderblock102seriallysuppliesthe（Y,U,V）componentsofacurrentframeoftheframesequencetoanadder/subtractor104.Adder/subtractor104isalsoadaptedtoreceiveasignalfrommotioncompensationblock106viaselector108.Ifselector108isintheupperposition,thenintra-framecodingisusedinwhichcaseanullsignal（i.e.,0）issuppliedtoadder/subtractor104.Ontheotherhand,ifselector108isinthelowerposition,theninter-framecodingisused.Adder/subtractor104generatesasignalthatcorrespondstothedifferencebetweenthevideodatasuppliedbyframereorderblock102andthatsuppliedbyselector108.

Thesignalgeneratedbyadder/subtractor104issuppliedtoadiscretecosinetransform（DCT）block110whoseoutputsignalisquantizedbyaquantizer112.Thequantizedsignalgeneratedbyquantizer112isthenencodedbyvariable-lengthcoder（VLC）114.ThesignalencodedbyVLC114isthenstoredinbuffer116,whichinturn,suppliestheencodedvideobitstreamtoavideodecoder（notshown）.

Thesignalgeneratedbyquantizer112isinverselyquantizedbyaninversequantizer118andissubsequentlydeliveredtoaninverseDCT（IDCT）120.IDCT20performsaninverseDCTfunctiononthesignalitreceivesandsuppliesthatsignaltoadder122.Adder122addsthesignalitreceivesfromselector108tothesignalitreceivesfromIDCT120andstorestheaddedresultinframememory124forfutureretrieval.Thesignalstoredinframememory124onlyincludesthelumacomponentofacurrentframeandisadaptedtoserveasareferenceframeformotionestimationandcompensationoffutureframes.

Amotionestimator128receivesthesignalstoredinframememory124andthesignalsuppliedbyframereorderblock102togenerateamotionvectorsignalthatissuppliedtomotioncompensationblock106.Onlythelumacomponentsofthecurrentframe—assuppliedbyframereorderblock102—andthereferenceframe—assuppliedbyframememory124—arereceivedandusedbymotionestimator128togenerateamotionvector.Themotionvectorgeneratedbymotionestimator128issuppliedtomotioncompensator106.Motioncompensator106,inturn,compensatesforthemotionofthesignalitreceivesfromframememory124usingthemotionvectorsignalthatitreceivesfrommotionestimator128.Theoutputsignalgeneratedbymotioncompensator106isamotion-compensatedsignalofacurrentframeandservesasthereferenceframeforthenextincomingframewheninter-frameencodingisused.

Theremaybeoccasionswhenareferenceframeisnotrequired.Forexample,noreferenceframeisrequiredwhenanewvideosequenceisreceivedbysystem100.Similarly,thereisnoneedforareferenceframewhenprocessingthefirstframeofanewscene.Toaccommodatesituationswherenoreferenceframeisneeded,selector108isprovidedwithanupperposition.Whenplacedintheupperposition,anullsignal（i.e.,0）istransferredtosubtracted104.

Conventionalluminance-onlybasedmotionestimationandcompensationsystems,suchassystem100,failtoreflectthetruemovementofanobjectinavideosequence.Thisfailureresultsinnoticeablecolor-motionartifacts.FIG.2isanexemplarydiagramshowingcolor-motionartifactsstemmingfromfailuretodetectthemotionofacolorobject.InFIG.2,auniformgrayarea200providesabackgroundtotwosyntheticcolorpatches,namelyaredcolorpatch210andagreencolorpatch220.Bothredandgreencolorpatches210and220producethesameluminancelevelasgraybackground200.Bothcolorpatches210and220arealsomovinginfrontofgraybackground200.

AssumethatinFIG.2,thecolorconversionrecommendedbytheITU-RstandardBT.709isused,asshownbelow:

Y=0.715G+0.072B+0.213R.

Assumefurtherthatgraybackground200has（R,G,B）colorcomponentsof（40,40,40）resultinginalumacomponent（i.e.,intensitylevel）of40.Assumefurtherthatredcolorpatch210andgreencolorpatch220haverespective（R,G,B）colorcomponentsof（188,0,0）and（0,56,0）.Consequently,inaccordancewithequation

（1）above,bothredcolorpatch210andgreencolorpatch220havethesameluminancelevel（i.e.,40）asthegraybackground200.Therefore,conventionalluminance-onlybasedmotionestimationandmotioncompensationsystems,suchassystem100ofFIG.1,failtodetectthemovementofredandgreencolorpatches210and220relativetograybackground200.Thisfailureresultsinnoticeabledifferencesinthequantizationerrorsofthechromacomponents—particularlyforthegraybackground—therebycreatingcolormotionartifacts.

Oneknownmethodforovercomingtheproblemsassociatedwithluminance-onlymotionestimationandcompensationsystemsisdescribedinU.S.Pat.No.5,544,263andwhichinvolvesperformingmotionestimationforeachofthecolorcomponents（Y,U,V）.Thisthree-motion-vectormethod,whileachievingagoodmatchforeachofthecolorcomponents,substantiallyincreasesthecomputationalintensity.Theincreaseincomputationalintensity,increasesboththecostofthesystemaswellasthebandwidthrequirementfortransmittingthethreemotionvectors.This,inturn,limitstheavailablebitratetoencodethemotion-compensatedinter-framedifferences.Therefore,althoughthemethodasdescribedinU.S.Pat.No.5,544,263hasanimprovedcolormotionartifactrejection,becauseitprovideslessbandwidthtoencodetheinter-framedifferences,itislessimmunetogranularnoise.

Anotherknownmethodforovercomingtheproblemsassociatedwithluminance-onlymotionestimationandcompensationsystemsisdescribedinU.S.Pat.No.5,544,263andwhichinvolvesusingallcolorcomponentswhenmatchingblocks.Inaccordancewiththismethod,block-matchingforeachmacroblockisperformedonlyonce.However,thistechniquemayrequire50%morecomputationthandotheluminance-onlymotion-estimationandcompensationsystems.This50%increaseincomputationincreasesthesystemcomplexityandcost.

Needcontinuestoexistforimprovedcolormotionartifactdetectionandreductiontechniques.

BRIEFSUMMARYOFTHEINVENTION