Precisionarmmechanism.docx

Precisionarmmechanism.docx

《Precisionarmmechanism.docx》由会员分享，可在线阅读，更多相关《Precisionarmmechanism.docx（11页珍藏版）》请在冰豆网上搜索。

Precisionarmmechanism

Digitalimagecodingusingrandomscanning

BACKGROUNDOFTHEINVENTION

Thepresentinventionrelatestodigitalimagecodingofvideosignals.

Theimmensepopularityofcompactdigitalaudiodiscscanbepartiallyattributedtothehighsoundqualityandfreedomfromdeteriorationoftherecordingduetoagingorrepeatedplayingsorfromenvironmentalfactorssuchasdust.Itispossibletoenjoysimilaradvantagesfromthedigitalencodingofvideosignalsaswell.However,realtimecodinganddecodingofavideosignalrequiresmuchmoresignalbandwidththandoesanaudiosignal,currentlyavailabletransmittingandrecordingmediaprovidesinsufficientstoragecapacityforvideosignals.Itisnecessary,therefore,tousesomeformofsignalcompressionordigitalbitreductiontechniqueinordertorecorddigitally-encodedvideo.Anumberofsuchbitreductionorsignalcodingmethods,thatwillenabletherecordingoffull-motiondigitalimages,havebeenproposedinrecentyears.

Becausetheavailabledatarateinpresenttransmissionchannelsandstoragemediaislimited,theseknowncodingmethodsacttoreducethedatarateofthevideosignal.Forthepurposeofstoringdigitally-encodedfullmotionimagesonopticalstoragemedia,suchasCD-ROMorMOD（MagnetoOpticalDisc）,forexample,amaximumvideodatarateofabout1.2Mbits/secondisallowed.

Oneexampleofasystemfordigitallycodinganddecodingvideosignalsisdisclosedinanarticleentitled"SceneAdaptiveCoder",W.H.Chen,W.K.Pratt,IEEETransactionsonCommunication,Vol.COM-32,No.3,issuedinMarch,1984inwhichsceneadaptivecoding/decodingofvideosignalsisdescribed.Thedescribedcodingtechniqueutilizesdiscretecosinetransform（DCT）,avariablethresholdfortheDCTcoefficients,quantizationtechniques,Huffmancodingandabufferinthecoder.Thedecoderincludesabuffer,Huffmandecoder,thresholdaddingandinverseDCT.

Anothersuchcoding/decodingsystemisdisclosedin"ComparisonBetweenProgressiveandInterlacedScanningforaFutureHDTVSystemwithDigitalDataRateReduction",D.Westerkamp,H.Peters,Proceedingsofthe2ndInternationalWorkshopofSignalProcessingforHDTV,l'Aquila,Feb.29-Mar.2,1988.Thedescribedsystemcontainsamotionadaptiveintrafield/interframecodingprocessor,inwhichthedigitizedvideodataistransformedbymeansofDCTtoobtainthequantizingcoefficients.

InthepreviouslydescribedsystemsthefullnessofthebuffercontrolsthethresholdandthequantizerstepsizefortheDCTcoefficients.Bothsystemsguaranteethatthebufferwillhavenooverfloworunderflowwithinagivenframe.Thiscausesavaryingquantizerstepsizeandaconsequentlyvaryingimagequalitywithineachframe.

InU.S.Pat.No.4,785,349,differentregionsofavideoframearecodedinadifferentmanner.Thismayalsocauseavaryingimagequalitywithineachframe.

Fromthedescriptionofthepreviouslydescribedsystems,itisknownthat,foragivencodingscheme,themoreablockofvideodataisquantized,thelessbitsarerequiredforcoding.Itisalsoknownthatthequantizationlevelmaybedeterminedbytheimage,orvideosignal,content.

TwocurrentlyknownmethodsmaybeusedtocomputethelevelofquantizationrequiredtocompressthedatarateofavideosignalwithinanavailablenumberofbitsN.Thefirstmethodcomputesathreshold/quantizationstepsizesothataframecanbecodedinapproximatelyNbits.Thequantizeristhenappliedwiththisuniformstepsizetothewholeframe.Thesecondmethodscansblocksofvideodatabyrowsandcolumnsinordertocomputealocalthreshold/quantizationstepsize.

Thefirstmethodallowsaspatiallyuniformquantizationstepsizetobeappliedtothewholeimage.However,computingtherightlevelofquantizationinordertocodetheimagetoascloseaspossibletoNbitsisverydifficult.Thesecondmethodallowsthecomputationofaquantizationstepsizeonablockbyblockbasis.However,thismethoddoesnotprovideauniformlevelofquantizationinthespatialdomain;thereforetheimagequalitymaybedifferentfortheblocksofeachframe.

SUMMARYOFTHEINVENTION

Inaccordancewithanaspectofthepresentinvention,amethodofencodingdigitalimagesequencescomprisesthestepsofgeneratingblocksofpixelsfromaframeofdigitalvideoinformation.Theblocksofpixelsarescannedinapseudo-randommannerandtransformedtoobtaincorrespondingblocksofcoefficients.Theblocksofcoefficientsarestoredinabuffer.Eachblockofcoefficientsareindividuallyquantizedbytheuseofaquantizationlevelthatiscontrolledbythefullnessofthebuffer.

Theadvantagesofthepresentinventionwillbecomemoreapparentfromthefollowingdetaileddescriptionwhentakeninconjunctionwiththeaccompanyingdrawing.

BRIEFDESCRIPTIONOFTHEDRAWING

Inthedrawing,FIG.1isablockdiagramofoneembodimentofacoderconstructedinaccordancewithanaspectofthepresentinvention;

FIG.2isablockdiagramofoneembodimentofadecoderconstructedinaccordancewithanaspectofthepresentinvention;

FIG.3isarepresentationofrandomlyscannedblocksofpixelswithinavideoimage;

FIG.4illustratesanexampleofaquantizationmatrixforablockofpixels;

FIG.5illustratesarepresentativepixelscanningmatrix;and

FIG.6isagraphillustratingtheoperationofaquantizerinaccordancewiththepresentinvention.

DETAILEDDESCRIPTION

Inordertoobtainavideoimageofmediumresolution,apixelmatrixoftheorderof704by240pixelsisnecessary.ThevideoinformationisillustrativelyencodedsuchthattheluminancecomponentYisencodedforeachofthepixels,whereasthecolorcomponentsUandVarecodedintheformofmacro-pixels,withamacro-pixelbeingmadeupof4luminancepixels.

Eachvideoframeisdividedintomacro-blocksof16by16luminancepixels.Eachmacro-blockiscomposedof4blocksof8by8pixelsfortheluminancecomponentYandoneblockof8by8macro-pixelsforeachofUandV.Asinglenon-encodedimagecomprising704by240pixelsconsistsof（704/16）×（240/16）=660macroblocks,whichequals（660×6）or3960blocks,equivalentto（3960×64）or253440bytes.

Aframeofthissizewouldoccupyabout250KByteinanuncodedformat,whereastheavailablebandwidthof1.2Mbits/secwouldonlyallowabout（1.2Mbit/s）/（8×30/s）=5KByteperframe,whichcorrespondstolessthan1/2bitperpixel.Acompressionrateofabout1/50isthereforerequiredtobeachieved.

InFIG.1adigitalimagesource11isconnectedtoarandomaccessmemory（RAM）12whichstoresthevideoinformationofaframeidentifiedasn.RAM12receivesblockaddresseswithinacurrentimagefromapseudo-randomblockaddressgenerator10.Eachblockofpixelsisthensenttoasubtractor141andtoamotioncompensationcircuit17.

Inmanyinstances,thevideoinformationcontainedinthepreviousframen-1isverysimilartothevideoinformationcontainedinthepresentorcurrentframen,onlythedifferencesbetweenframesn-1andnareencoded.Thistechniqueiscalledinterframecoding（delta-coding）andresultsinarelativelylowdatarate.

However,insomecasesitisnecessarytoencodeaframebyitself,suchaswhenthereisa"cut"orscenetransitionintheimagesequence.Thistechniqueiscalledintraframecoding.Itisalsonecessarytouseintraframecodingtoallownon-sequentialdecodingwhenitisdesirabletostartdecodingfromvariouspointswithinanimagesequencestoredonaCD,suchaswouldberequiredforrandomaccesstovideoinformationonadisc.Intraframecodingdoesresultsinahigherdatarate,however.

Whenthereismotionintheimagesequence,interframecodingmayyieldpoorqualitycodedimages.Inthosesituations,themotionintheimagesequenceshouldbedetected.These"motion"differencesbetweencorrespondingelementsinframesn-1andnarethenusedtominimizethecodingerroroftheinterframecoding,therebyleadingtoimprovedimagequality.Thepurposeofmotioncompensationistobuildanimageascloseaspossibletoframenbyusingthedetectedmotionelementsfromframen-1.

Thisprocessisperformedinthemotioncompensationcircuit17onapixelblockbasis.AblockB（n,i,j）fromframen,locatedatframecoordinates（i,j）,forexample,isevaluatedbycopyingtheblockB'（n-1,i+x,j+y）locatedatframecoordinates（i+x,j+y）inaframememory（RAM）142whichcontainsframen-1.AmotioncompensationvectorV（n,x,y）,representingtheapparentmotionoftheimageatframecoordinates（i,j）onthescreen,isgenerated.Whenusing8by8pixelblocksformotioncompensation,themotioncompensationvectorsmayhaveaprecisionof±1pixelandarangeof±16pixels.

Onceagoodapproximationofframenhasbeenobtainedfromframen-1byusingmotioncompensation,theresidualpixeldifferencesremaintobeencoded.Anestimationofthedifferencebetweencorrespondingblocksofframesn-1andnisdeterminedandadecisionismadewithrespecttoeachblockastowhethermotioncompensatedinterframecodingorintraframecodingyieldsbetterimagequalityinconnectionwithagivendatarate.Thisdecisionismadebyerrorevaluationcircuit18bywayofevaluationofaquadraticerrorfunction.Theequationofthequadraticerroris:

##EQU1##whereiiandjjarecoordinateswithintheblocksB（n,i,j）andB'（n-l,i+x,j+y）,respectively,ofpixelsPandP',respectively.E（B,B'）isanestimationofthedifferencebetweenblockB（n,i,j）offramenandblockB'（n-l,i+x,j+y）offramen-1.B（n,i,j）isthetargetblock（notcoded）andB'（n-l,i+x,j+y）isablockobtainedfromframen-1afterinterframecodingTheencodingo