文献翻译——数字图像处理方法的研究

1外文文献资料Part1TheresearchofdigitalimageprocessingtechniqueInterestindigitalimageprocessingmethodsstemsfromtwoprincipalapplicationareas:improvementofpictorialinformationforhumaninterpretation;andprocessingofimagedataforstorage,transmission,andrepresentationforautonomousmachineperception.Thischapterhasseveralobjectives:(1)todefinethescopeofthefiledthatwecallimageprocessing;(2)togiveahistoricalperspectiveoftheoriginsofthisfiled;(3)togiveanideaofthestateoftheartinimageprocessingbytheexaminingsomeoftheprincipalareainwhichitisapplied;(4)todiscussbrieflytheprincipalapproachesusedindigitalimageprocessing;(5)togiveanoverviewofthecomponentscontainedinatypical,general-purposeimageprocessingsystem;(6)toprovidedirectiontothebooksandotherliteraturewhereimageprocessingworknormallyisreporter.1.1WhatIsDigitalImageProcessing?Animagemaybedefinedasatwo-dimensionalfunction,f(x,y),wherexandyarespatial(plane)coordinates,andtheamplitudeofatanypairofcoordinates(x,y)iscalledtheintensityorgrayleveloftheimageatthatpoint.Whenx,y,anddigitalimage.Thefiledofdigitalimageprocessingreferstoprocessingdigitalimagebymeansofadigitalcomputer.Notethatadigitalimageiscomposedofafinitenumberofelements,eachofwhichhasaparticularlocationandvalue.Theseelementsarereferredtoaspictureelements,imageelementsofadigitalimage.WeconsiderthesedefinitionsinmoreformaltermsinChapter2.Visionisthemostadvancedofoursenses,soitisnotsurprisingthatimagesplaythesinglemostimportantroleinhumanperception.However,unlikehumanwhoarelimitedtothevisualband2oftheelectromagnetic(EM)spectrum,imagingmachinescoveralmosttheentireEMspectrum,rangingfromgammatoradiowaves.Theycanoperateonimage.Theseincludeultrasound,electronmicroscopy,andcomputer-generatedimages.Thus,digitalimageprocessingencompassesawideandvariedfiledofapplication.Thereisnogeneralagreementamongauthorregardingwhereimageprocessingstopsandotherrelatedareas,suchasimageanalysisandcomputervision,start.Sometimesadistinctionidmadebydefiningimageprocessingasadisciplineinwhichboththeinputandoutputofaprocessareimages.Webelievethistobealimitingandsomewhatartificialboundary.Forexample,underthisdefinition,eventhetrivialtaskofcomputingtheaverageintensityofanimage(whichyieldasinglenumber)wouldnotbeconsideredanimageprocessingoperation.Ontheotherhand,therearefieldssuchascomputevisionwhoseultimategoalistousecomputertoemulatehumanvision,includinglearningandbeingabletomakeinferencesandtakeactionsbasedonvisualinputs.Thisareaitselfisabrandofartificialintelligence(AI)whoseobjectiveistoemulatehumanintelligence.ThisfiledofAIisinitsearlieststagesofinfancyintermsofdevelopment,withprogresshavingbeenmuchslowerthanoriginallyanticipated.Theareaofimageanalysis(alsocalledimageunderstanding)isinbetweenimageprocessingandcomputervision.Therearenoclear-cutboundariesinthecontinuumfromimageprocessingatoneendtocomputervisionattheother.However,oneusefulparadigmistoconsiderthreetypesofcomputerizedprocessesisthiscontinuum:low-,mid-,andhigh-everprocesses.Low-levelprocessesinvolveprimitiveoperationsuchasimageprocessingtoreducenoise,contrastenhancement,andimagesharpening.Alow-levelprocessischaracterizedbythefactthatbothitsinputandoutputareimages.Mid-levelprocessischaracterizedbythefactthatitsinputsgenerallyareimages,butitsoutputisattributesextracted3fromthoseimageanalysis,and,atthefarendofthecontinuum,performingthecognitivefunctionnormallyassociatedwithvision.Basedontheprecedingcomments,weseethatalogicalplaceofoverlapbetweenimageprocessingandimageanalysisistheareaofrecognitionofindividualregionsorobjectsinanimage.Thus,whatwecallinthisbookdigitalimageprocessingencompassesprocesseswhoseinputsandoutputsareimagesand,inadditionencompassesprocessesthatextractattributesfromimages,uptoandincludingtherecognitionofindividualcharacters,describingthecharactersinaformsuitableforcomputerprocessing,andrecognizingthoseindividualcharactersareinthescopeofwhatwecalldigitalimageprocessinginthisbook.Makingsenseofthecontentofthepagemaybeviewedasbeinginthedomainofimageanalysisandevencomputervision,dependingonthelevelofcomplexityimpliedbythestatement“makingcense”.Aswillbecomeevidentshortly,digitalimageprocessing.Aswehavedefineditisusedsuccessfullyinabroadrangofareasofexceptionalsocialandeconomicvalue.Theconceptsdevelopedinthefollowingchaptersarethefoundationforthemethodsusedinthoseapplicationareas.1.2TheOriginsofDigitalImageProcessingOneofthefirstapplicationifdigitalimageswasinthenewspaperindustry,whenpicturewerefirstsentbysubmarinecablebetweenLondonandNewYork.IntroductionoftheBartlanecablepicturetransmissionsystemintheearly1920sreducedthetimerequiredtotransportapictureacrosstheAtlanticfrommorethanaweektolessthanthreehours.Specializedprintingequipmentcodedpicturesforcabletransmissionandthenreconstructedthematreceivingend.Someoftheinitialprobleminimprovingthevisualqualityoftheseearlydigitalpicturewererelatedtotheselectionofprintingproceduresandthedistributionofintensitylevels.Theprinting4methodabovewasabandonedtowardtheendof1921infavorofatechniquebasedonphotographicreproductionmadefromtapesperforatedatthetelegraphreceivingterminal.TheearlyBartlanesystemswerecapableofcodingimagesinfivedistinctlevelofgray.Thiscapabilitywasincreasedto15levelsin1929.Duringthisperiod,introductionofasystemfordevelopingafilmplatevialightbeamsthatweremodulatedbythecodedpicturetapeimprovedthereproductionprocessconsiderably.Althoughtheexamplesjustcitedinvolvedigitalimages,theyarenotconsidereddigitalimageprocessingresultsinthecontextofourdefinitionbecausecomputerwerenotinvolvedintheircreation.Thus,thehistoryofdigitalprocessingisintimatelytiedtothedevelopmentofthedigitalcomputersofsupportingtechnologiesthatincludedatastorage,display,andtransmission.5IntroductiontoMATLABMATLABisaspecial-purposecomputerprogramoptimizedtoperformengineeringandscientificcalculations.Itstartedlifehasaprogramdesignedtoperformmatrixmathematics,butovertheyearsithasgrownintoaflexiblecomputingsystemcapableofsolvingessentiallyanytechnicalproblems.TheMATLABprogramimplementstheMATLABprogramminglanguageandprovidesanextensivelibraryofpredefinedtheMATLABprogramminglanguageandprogrammingtaskseasierandmoreefficient.ThisbookintroducestheMATLABlanguageandshowshowtouseittosolvetypicaltechnicalproblems.MATLABisahugeprogram,withanincrediblyrichvarietyoffunction.EventhebasicversionofMATLABwithoutanytoolkitsismuchricherthanothertechnicalprogramminglanguages.Therearemorethan1000functioninthebasicMATLABproductalone,andthetoolkitextendthiscapabilitywithmanymorefunctionsinvariousspecialties.ThisbookmakesnoattempttointroducetheusertoallofMATLABsowntoolstolocatethecorrectfunctionforaspecificpurposefromtheenormouschoiceavailable.AdvantagesofMATLABMATLABhasmanyadvantagescomparedwithconventionalcomputerlanguagesfortechnicalproblemsolving.Amongthemarethefollowing:1.EaseofuseMATLABisaninterpretedlanguage,likemanyversionsofbasic,itisveryeasytouse.Theprogramcanbeusedasascratchpadtoevaluateexpressionstypedatthecommandline,oritcanbeusetoexecutelargeprograms.Programsmaybeeasilywrittenandmodifiedwiththebuilt-inintegrateddevelopmentenvironment,anddebuggedwiththeMATLABdebugger.Becausethelanguageissoeasytouse,itisidealforeducationaluse,andfortherapidprototypingofnewprograms.6Manyprogramdevelopmenttoolsareprovidedtomaketheprogrameasytouse.Aworkspacebrowser,andextensivedemos.2.PlatformindependenceMATLABissupportedonmanydifferentcomputersystems,providingalargemeasureofplatformindependence.Atthetimeofthiswriting,thelanguageissupportedonwindows9x/NT/2000andmanydifferentversionsofUNIX.Programswrittenonanyplatformwillrunonalloftheotherplatforms,anddatafileswrittenonanyplatformmaybereadtransparentlyonanyotherplatformmaybereadtransparentlyonanyotherplatforms.Asaresult,programswritteninMATLABcanmigratetonewplatformwhentheneedsoftheuserchange.3.PredefinedFunctionsMATLABcomescompletewithanextensivelibraryofpredefinedfunctionsthatprovidetestedandprepackagedsolutionstomanybasictechnicaltasks.Forexample,supposethatyouarewritingaprogramthatmustcalculatethestatisticsassociatedwithaninputdataset.Inmostlanguages,youwouldneedtowriteyourownsubroutinesorfunctionstoimplementcalculationssuchasthearithmeticmean,standarddeviation,median,andsoforth.TheseandhundredofotherfunctionarebuiltrightintotheMATLABlanguage,makingyourjobmucheasier.InadditiontothelargelibraryoffunctionsbuiltintothebasicMATLABlanguage,special-purposetoolboxesareavailabletohelpsolvecomplexproblemsinspecificareas.Forexample,ausercanbuystandardtoolboxestosolveprobleminSignalProcessing.ControlSystem,Communications,ImageProcessing,andNeuralNetworks,amongmanyothers.Thereisalsoanextensivecollectionoffreeuser-contributedMATLABprogramsthataresharedthroughtheMATLBWebsite.4.Device-IndependentPlottingUnlikemostothercomputerlanguages,MATLABhasmanyintegral7plottingandimagingcommands.TheplotsandimagescanbedisplayedonanygraphicaloutputdevicesupportedbythecomputeronwhichMATLABisrunning.ThiscapabilitymakesMATLABanoutstandingtoolforvisualizingtechnicaldata.5.GraphicalUserInterfaceMATLABincludestoolsallowaprogrammertointeractivelyconstructagraphicaluserinterface(GUI)forhisorherprogram.Withthiscapability,theprogrammercandesignsophisticateddataanalysisprogramsthatcanbeoperatedbyrelativelyinexperiencedusers.6.GraphicsprocessingMATLABisgeneratedfromthedatehasthefunctionofconvenientdatavisualization,thevectorandmatrixgraphicalpresentation,andcanbeannotatedandprintgraphics.Mappingofhighlevelincluding2Dand3Dvisualization,imageprocessing,animationandgraphicexpression.Canbeusedinscientificcomputingandengineeringdrawing.ThenewversionoftheMATLABonthegraphicsprocessingfunctionwasgreatlyimprovedandimproved,sothatitnotonlyhasthegeneraldatavisualizationsoftware(e.g.2Dcurveand3Dsurfaceprocessingandsoon)moreperfect,andforsomeothersoftwarenotsomefunctions(suchasgraphicsilluminationprocessing,colorfourdimensionaldataprocessingandperformance),MATLABalsoshowtheprocessingabilityoutstanding.Atthesametime,somespecialvisualrequirements,suchasgraphicaldialogue,MATLABalsohasthecorrespondingfunctions,ensurethatthedifferentlevelsofuserrequirements.InadditiontothenewversionofMATLABalsofocusesonthegraphicaluserinterface(GUI)productionweregreatlyimproved,thespecialrequirementsofuserscanbesatisfiedonthisaspect.MATLABhasdevelopedapowerfulsetofmodulesandtoolboxformanyspecializedfield.Generallyspeaking,theyareexpertsfromspecific8areasofdevelopment,theusercanlearn,applicationandevaluationofdifferentuseddirectlywithouttheneedtowriteyourowncode.Fields,suchasdataacquisition,databaseinterface,probabilityandstatistics,splinefitting,optimizationalgorithm,partialdifferentialequation,neuralnetwork,waveletanalysis,signalprocessing,imageprocessing,systemidentification,controlsystemdesign,LMIcontrol,robustcontrol,modelpredictive,fuzzylogic,financialanalysis,maptool,nonlinearcontroldesignrealtime,rapidprototypingandsemiphysicalsimulation,embeddedsystemdevelopment,fixed-pointsimulation,DSPandcommunications,electricpowersystemsimulationtoolbox(Toolbox),inthefamilyhasitsownaspaceforoneperson.7.StrongprocessingabilityMATLABisacollectioncontainingalargeamountofcalculationalgorithm.Ithasmorethan600projectsintheuseofthemathematicaloperationfunction,cancalculatethefunctionrealizationofvarioususersforconvenient.Thefunctionofthealgorithmisthelatestresearchachievementsofscientificandengineeringcomputing,andthroughavarietyofoptimizationandfault-tolerantprocessing.Undernormalcircumstances,itcanbeusedtoreplacetheunderlyingprogramminglanguage,suchasCandC+.Inthecomputationalrequirementsunderthesameconditions,theuseofMATLABprogrammingworkloadwillbegreatlyreduced.TheseMATLABfunctionsrangingfromthemostsimplethemostbasicfunctionssuchasmatrix,complexfunctionfeaturevector,theFIR.Functioncansolvetheproblemgenerallyincludesmatrixandsolvinglinearequations,differentialequationsandpartialdifferentialequationgroupsolution,symboliccomputation,FuLiyetransformandthestatisticalanalysisofthedata,theoptimizationprobleminengineering,thesparsematrix,apluralityofvariousarithmetic,trigonometricfunctionsandotherelementarymathematicaloperations,9amultidimensionalarrayoperationandmodeling,dynamicsimulation.DisadvantagesofMATLABMATLABhastwoprincipaldisadvantage.Thefirstisthatitisaninterpretedlanguage,andthereforecanexecutemoreslowlythancompiledlanguages.ThisproblemcanbemitigatedbyproperlystructuringtheMATLABprogrambeforedistributionandgeneraluse.Theseconddisadvantageiscost:AfullcopyofMATLABis5to10timesmoreexpensivethanaconventionCorFortrancompiler.Thisrelativelyhighcostismorethanoffsetbythereducedtimerequiredforanengineerorscientisttocreateaworkingprogram,soMATLABiscost-effectiveforbusinesses.However,itistooexpensiveformostindividualstoconsiderpurchasing.Fortunetely,thereisalsoaninexpensiveStudentEditionofMATLABisessentiallyidenticaltothefulledition.Withtheintroductionofbranchesandloops,ourprogramsaregoingtobecomemorecomplex,anditwillgeteasiertomalemistakes.Tohelpavoidprogrammingerrors,wewillintroduceaformalprogramdesignprocedurebasedonthetechniqueknownastop-downdesign.Wewillalsointroduceacommonalgorithmdevelopmenttoolknownaspseudocode.IntroductionToTop-DownDesignTechniquesIfyouworkinafactoryengineer,inordertosolvesomeproblem,youaretowriteaprogram.Howdoyoustart?Whenanewproblemis,ourheartscomeverynaturallygeneratesaidea:Immediatelysatinfrontofthecomputertostartprogramming,insteadofwastingalotoftimethinkingabouttheproblemweneedtosolveiswhat?Thisunrealisticideatomakesomeverysmallprogramislikelytobesuccessful.Butinpractice,theproblemmaybeverylarge,programverificationandprogrambythismethodwillgetintotrouble.Foralargeprogram,thinkbeforeyouwritethecodeyouallyouneedtofacetheproblemandsolvingmethod.Inthis10section,wewillintroducetheprogrammingstepsofformal,andthenapplythisproceduretowriteusinalllargeapplicationafteruse.Forsomesimpleexamplesweencountered,thisseemsabitsuperfluous.Butwhenwewanttosolvetheproblemofincreasing,thisstepwillbecomeextremelyimportant.Whenwehaventgraduation,likeaprofessorsaid:programmingisverysimple,becauseIknowitisdifficultintheprogrammingprocess.Whenwehaveleftschool,engagedinlarge-scalesoftwareprogramminginthefactory,Ideeplyunderstandwhathesaid.IfoundthatmostofthedifficultiesIencounteredintheworkistheneedtosolvetheproblemofunderstanding.Onceyouunderstandtheproblem,youwillgetthisproblemisdecomposedintomanysmallpieces.Andlecturerprogrammingtop-downmethodournormalprogramming.1.ClearlystatetheproblemthatyouaretryingtosolveProgramsareusuallywrittentofillsomeperceivedneed,butthatneedmaynotearticulatedclearlybythepersonrequestingtheprogram.2.DefinetheinputsrequiredbytheprogramandtheoutputstobeproducedbytheprogramForinputandoutput,onlythisnewprograminordertoadapttothewholeprocessplan.Coefficientequationinthisexamplemayhavepre-existingorder,ournewproceduremustbeabletoreadtheminorder.Similarly,alsoneedtoproducetheprogramneededresults,namelytheoutput,wealsoinacertainformatprint3.Designthealgorithmthatyouintendtoimplementintheprogram11StarttheproblemyouaretryingtosolveDefinerequiredinputsandoutputsTesttheresultingMATLABprogramConvertalgorithmintoMATLABstatementsDefinethealgorithmDecompositionstepwiserefinementThealgorithmistofindtheanswertoastep-by-stepprocedureforaproblem.Playaroleinthisstagethetop-downprogrammingmethod.Thedesignerstartedprogramminglogicpartitionofthisproblem,itisgraduallybrokendownintoasubjob.Thisprocessiscalleddecomposition(decomposition).Iftheworkisstillrelativelylarge,thedesigneralsoletitintosmallerblocks.Thisprocesswillbecontinuedproblemisdecomposedintomanysimpleandeasytounderstandsmall.Theproblemisdecomposedintosmallpieces,eachpieceshouldbefurtherrefinement,thisprocessiscalledstepwiserefinement,inthisprocess,designersbegantothisschoolfastcodeinthedescription,andthenstartthefunctiondefinitionstepbystepinneedofmore,moreandmorespecific,untilhecanbeconvertedtoMATLABstatement.Stepwiserefinementprocess,wewillstartfinish12usethepseudocodewillinthenextsectiontointroduce.Indevelopingthealgorithm,thismethodisveryuseful,ifthedesignerofthesestepsinunderstandingthereal,hewilldecomposeandstepwiserefinementoftheproblem.4.TurnthealgorithmintoMATLABstatementsIftheprocessofdecompositionandrefinementhasbeensuccessfullycompleted,sothisstepwillbeabnormalsimple.AllprogrammerswilltellthepseudocodeasentenceintotheappropriateMATLABcode5.TesttheresultingMATLABprogramThisstepistherealobstacle.Firstofall,everypartoftheprogramwillbeseparatedetection,ifpossible,thewholeprocesswouldbedetectedagain.Inourtestprocedureis,wemustprovethatalllegitimateinputvaluescanbenormaloperation,testingproceduresusingastandardinputvalues,seeitdienotproducevalue.Ifimplementedinaprogramalgorithmcontainsadifferentbranch,youmusttesteachbranch,toproducethecorrectanswertoensure.Largeprogrambeforedeliveryofpublicuse,mustcarryonaseriesoftests.Thefirststepintestingissometimesreferredtoasaunittest.Inunittesting,testingprogramsubroutinewillbeindependenttoprovehimright.Whentheunittestisfinished,theprogramwillbeaseriesofcombination,theindependentsubprogramsjointlyproduceafinalprogram.Thefirststepintheprocessofjointusuallyincludedonlyafewsubroutine.Bycombiningthesesubroutines,oftenwithasubroutineorfunctionrelationshipbetweencombination.Thecombinationofaseriesofprocess,everymistakewillbefoundandcametocorrectbeforethenextcombination.13subtasksvalidatedseparatelyAsmanyasnecessaryAsmanyasnecessaryAsmanyasnecessarysubtaskscombinedintoasingleprogramworstbugsfixedUnittestingofindividualsubtasksAlphareleaseAseriesofcombination(addingsubtaskstotheprogramAlphareleaseminorbugsfixedFinishedprogramstartFigure214中文翻译稿数字图像处理方法的研究数字图像处理的研究源于两个主要领域：奇异是为了便于人们分析二对图像信息进行该井；其二是为了使机器自动理解而对图像数据进行存储、传输及显示。1.1数字图像处理的概念一幅图像可以定义为一个二维函数f（x，y),这里x和y是空间坐标，而在任何一对空间坐标f（x，y）上的幅值f称为该点图像的灰度或强度。当x,y和幅值f为有限的、离散的数值时，称改点是由有限的元素组成的，每一个元素都有特定的位置和幅值，这些元素称为图像元素、画面元素或像素。像素是广泛用于表示数字图像元素的词汇。视觉是人类最高级的感知器官，所以毫无疑问图像在人类感知中扮演着最重要的角色。然而，人类感知只限于电磁波谱的视觉波段，成像极其则可覆盖几乎全部电磁波谱，从伽马射线到无线点拨。它们可以对非人类习惯的那些图像源进行加工，这些图像源保罗超声波、电子显微镜及计算机产生的图像。因此，数字图像处理涉及各种各样的应用领域。图像处理涉及的范畴或其他相关领域（例如，图像分析和计算机视觉）的界定在初创人之间并没有一致的看法。有时用处理的输入和输出内容都是图像这一特点来界定图像处理的范围。我们认为这一定义仅是人为界定和限制。例如，在这个定义下，甚至最普通的计算一副图像灰度平均值的工作都不能算作是图像处理。另一方面，有些领域（如计算机视觉）研究的最高目标是用