visual computing 视觉计算学习体会.docx
- 文档编号:29605103
- 上传时间:2023-07-25
- 格式:DOCX
- 页数:11
- 大小:185.47KB
visual computing 视觉计算学习体会.docx
《visual computing 视觉计算学习体会.docx》由会员分享,可在线阅读,更多相关《visual computing 视觉计算学习体会.docx(11页珍藏版)》请在冰豆网上搜索。
visualcomputing视觉计算学习体会
VisualComputing
Lasttwoweeks,wehavetakenthecourseVisualComputing.Thesubjectisveryinteresting,neotericandmeaty.ThroughtheExcellentpresentationoftheteachers,Iknowmoreaboutthissubjectandgetinterestedinit.
WhatisVisualComputing?
VisualComputingisanexcitingnewresearchareathatstudieshowtomakecomputersefficientlyperceive,process,andunderstandvisualdatasuchasimagesandvideos.Computervisionistheknowledgeontheuseofcameraandcomputertoobtaintherequiredinformationofthesubject.Figurativelyspeaking,thatisinstalledonyourcomputerontheeye(camera)andbrain(algorithm),allowthecomputertobeabletoperceivetheenvironment.
Thereisnodoubtthatthecomputercannotobtaintheinformationlikehumanfromthevisualmessageitget.Forthehuman,wecangetmuchinformationotherthansize,distance,color.Forexample,wecanjudgethefeelingoftheotherpersonjustthroughpettyactionorpettyfacialexpressionofthepeople.Butthecomputercannotdoiteasily.Itisjustnoteasyeasilyforthecomputertojudgethefeelingorsomeotherabstractthingsthroughthevisualvisionofcamera.Thus,theultimategoalforcomputersistoemulatethestrikingperceptualcapabilityofhumaneyesandbrains,oreventosurpassandassistthehumanincertainways.
Therearemanyfamousscientistsinthisfield.Amongthem,IthinkthefamousoneisD.Marr.Heisafamousscientistinthefieldofvisualcomputing,whohadputforwardmanytheoryaboutvisualcomputing.Hesaid,VisionResearch'sultimategoalistoclarifywhetherthevisualsystemishowtocompletethevisualtask.Andheconsideredthattheinformationprocessingbythenervoussystemandmachineissimilar.Thevisionisacomplexinformationprocessingtask,tograspthevarioususefulandmeaningfuloutsideworld,andtoexpressthem.Thistaskmustbeunderstoodatthreedifferentlevels,whichis:
a.Calculationtheory;b.Algorithm;c.Mechanism.Thisdivisionisnotverystrict,butifyoudonotcomplywiththeaboveclassification,thenthatisnotoneoragroupdescriptionisperfect.Inthevisualperception,eachofthethreelevelshasitsspecificlocation.Theyarebasicallyindependentofeachother.Thus,intheexplorationoftheoreticalissues,wemustbeusenewresearchmethods,andmakeitstrictlydistinguishedfromlawandmechanisms.
TheabovemethodputforwardbyD.Marrissostrong,itjustmakesthevisualinformationsciencesjumpintotherapiddevelopmentandgrowth,withwhichtheresultisthatitjustlikephysicsasapermanentnature.Becausetheyareasolidfoundationonthebasiclawsofphysicsandformulaicimageoftherealworld,thus,thislevelofvisualcomputingtheoryinthedevelopmentprocessmaybecomearealscience.D.Marr'sworkisfromthemainthemeofcalculationtothebasisofthetheoryofanalyzetheconcretedetailsofthemethodological.Heisjustagreatandexcellentscientist!
Comebacktotheconcretecontenttoourcourse,therearemanyinterestingcontent.Andatthebeginningofthecourse,theteacherfromItalywhonamedP.LECALLET,telltheknowledgeabouttheintroductiontoVideocodingstandardmotionestimation.Thoughhegotintothemajorpointquickly,whichmadeusalittlebitdifficultandconfusedtogetthepointofhim,wewasabsorbedbywhathesaidlittlebylittleandinterestingjokesofhim.
Firstly,welearntheuncompresseddigitalvideo:
bit-rate.Thebit-ratejustdescribesthetransferspeedbythenumberofbitstransferredinonesecond.Allofusknowthatthesmallestunitinthesystemofcomputerisbit.Anditisobviouslythatmachinecantransferredmoredatainunittimewithhighbit-rate.Wehavegotsomebit-rateofvideoforms.Wecangetthattheclearervideoformhasthehigherbit-rate.WeknowthattheHDisclearerthantheSD.
Thenforthecompresseddigitalvideo,itcanbedividedintotwotypologies:
lossyorloss-less.Forthelossycase,ithasonespecialcasewhichnamedvisuallyloss-less.Itisamazing.Whatismore,forthelossycompression,weneedtodefineacceptablevisualquality,whichmayvarywiththetargetedservices.
AllofuswillliketowatchtheHDmoviesnowfortheclearerimageandvisualfeelingofthisvideoform.Butforthelossycompression,itjustmakesgooduseofthecharacteristicsthatthehumanisnotsensitivetocertainfrequencycomponentsintheimageoracoustic,whichallowsacertainamountofinformationduringthecompressionloss.Althoughitcannotcompletelyrestoretheoriginaldata,thelossofpartofthetakeslittleeffecttotheunderstandingoftheoriginalimage,inreturnforamuchlargercompressionratio.Nowadays,lossycompressioniswidelyusedinthevoice,imageandvideodatacompression.Thecommonsound,imageandvideocompressionarebasicallylossy.Inmultimediaapplications,acommoncompressionmethodis:
PCM(PulseCodeModulationlossycompression),Predictivecoding,transformcoding,interpolationandextrapolation,statisticalcoding,vectorquantizationandsub-bandcoding,hybridcodingisawidelyusedmethodinrecentyears.Thevideoformmp3,divX,Xvid,jpeg,rm,rmvb,wma,wmvarealllossycompression.
Theadvantageofthecompressionisthatthefilesizeismuchsmallerthananyknownnondestructivemethod,whileatthesametimetomeettheneedsofthesystem.Lossycompressionfilewhentheuser,forexample,inordertosavedownloadtime,theunzipfileandtheoriginalfileinthedatabitlevelpointofviewmaybeverydifferent,butformostpracticalpurposes,thehumanearorthehumaneyecannotdistinguishbetweenthedifferencebetweenthem.Andwiththelossycompressiontechniques,certaindataisintentionallydeleted,thedataisalsonolongerbecanceledtorestore.
Forthelossycompressionandtheloss-lesscompression,weconsiderthattheperformancedependsonthreetheories:
bit-rate,ratedistortioncurveandtheMoore’slaw.
Therearetwomajormethodstorealizethecompression.Oneislossytransformcoding,whichcutthedataintosmallpieces,andconvertthemintoanewspace,thenquantizethem.Finallydoentropycodingtothequantizedvalues.Anotherisnamedpredictivecoding,thepreviousdata,andlaterdecodeddatauselossycompressiontopredictthesamplingofthecurrentsoundorimageframestogetthedataoferrorbetweenthepredictiondataandtheactualdataaswellasdothequantizationandcodingofsomeotherinformationtoreproducetheprediction.
Ininformationtheory,entropyencodingisaloss-lessdatacompressionschemethatisindependentofthespecificcharacteristicsofthemedium.Oneofthemaintypesofentropycodingcreatesandassignsauniqueprefix-freecodetoeachuniquesymbolthatoccursintheinput.Theseentropyencodersthencompressdatabyreplacingeachfixed-lengthinputsymbolwiththecorrespondingvariable-lengthprefix-freeoutputcodeword.Thelengthofeachcodewordisapproximatelyproportionaltothenegativelogarithmoftheprobability.Therefore,themostcommonsymbolsusetheshortestcodes.TwoofthemostcommonentropiesencodingtechniquesareHuffmancodingandarithmeticcoding.Iftheapproximateentropycharacteristicsofadatastreamareknowninadvance(especiallyforsignalcompression),thenasimplerstaticcodemaybeuseful.Thesestaticcodesincludeuniversalcodes(suchasEliasgammacodingorFibonaccicoding)andGolombcodes(suchasunarycodingorRicecoding).
FortheHuffmancoding,themajorpointistoavoidrepetition.AndtheHuffmancodingusuallyappearsasthebinarytree,forwhichtheendingpointsrepresentsthecodingletters,andtherootnodeisonbehalfofthebits.Thusforthecharacterswithhighoccurrencerate,weshouldletthecodeofittobeshorterthantheothers.Inthisway,theaveragelengthofexpectedstringswillbereducedsoastoachievethepurposeofcompression.
Andthen,fortheothermethodwhichnamedpredictivecoding,differentfromtheentropycodingabove,itisbasedonthecorrelationbetweeninformation.Thepredictivecodingisbasedonthecharacteristicsthattherearecertaincorrelationbetweenthediscretesignals,usingthelastoneormoresignalstopredictthenextsignaltobeencoded,thendothecodingwiththedifference(predictionerror)betweentheactualvalueandthepredictedvalue.Iftheforecastisaccurate,theerrorwillbeverysmall.Intherequiredconditionsofthesameaccuracy,itispossibletousefewerbitsforencoding,andtoachievethepurposeofthecompresseddata.
Thenwelearnthediscretecosinetransformation(DCT).Thediscretecosinetransformation(DCT)isatransformationrelatedtoFouriertransformation,whichissimilartothediscreteFouriertransform(DFT),butusingonlytherealnumber.DiscretecosinetransformationisequivalenttoadiscreteFouriertransformationtwiceofthelength,whichiscarriedoutononedualfunction(functionasarealdualfunctionoftheFouriertransformationisstillarealdual),somedeformationwhichneedstobeinputtedoroutputtedpositionthemobilehalfaunit.Whatismore,DCThaseightstandardtypes,fourofwhicharecommon.
ThroughtheDCT,wecantranslatethespatialdomaintofrequencydomain.ThereisaexampleofDCT:
Forthedigitalvideocompression,thegeneralprincipleisthattheperformanceofthecompressionisjustrelatedtotheremoveofsignalredundancy.Thegeneralprocessesofthecompressionaredothetransformationofthesource,thendothequantization,andusingtheentropycodingmethodtocode.Finallywecangetthecompresseddata.Forthevideocase,thereare3typesof
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- visual computing 视觉计算学习体会 视觉 计算 学习体会