数字图像处理与边缘检测英译中.docx

1、数字图像处理与边缘检测英译中DigitalImageProcessingandEdgeDetectionDigitalImageProcessingInterestindigitalimageprocessingmethodsstemsfromtwo principalapplica-tionareas:improvementofpictorialinformationforhumaninterpretation;and processing of image data for storage, transmission, and representationforau-tonomous ma

2、chineperception.Animagemaybedefinedasatwo,dimensionalfunction,f(x,y),wherexandyarespatial(plane)coordinates,andtheamplitude offatanypairofcoordinates(x,y) is called the intensityorgraylevelof the image at that point.Whenx,y,andtheamplitudevaluesoffareallfinite, discretequantities,we call theimageadi

3、gitalimage. Thefieldofdigitalimageprocessingreferstoprocessingdigitalimagesbymeansofadigitalcomputer.Notethatadigitalimageiscomposedofafinitenumberofelements,eachofwhichhasaparticularlocationandvalue.Theseelementsarereferredtoaspictureelements,imageelements,pels,andpixels.Pixelisthetermmostwidelyuse

4、dtodenotetheelementsofadigitalimage.Visionisthemostadvancedofoursenses,soitisnotsurprisingthatimagesplaythesinglemostimportantroleinhumanperception.However,unlikehumans,whoarelimitedtothe visual bandoftheelectromagnetic(EM)spec-trum, imaging machinescoveralmosttheentireEMspectrum, ranging fromgammat

5、oradiowaves.Theycan operate on imagesgeneratedbysourcesthathumans are not accustomedtoassociatingwithimages.Theseincludeultra-sound,electronmicroscopy,andcomputergeneratedimages.Thus, digital imageprocessingencompassesawideandvariedfieldofapplications.Thereisnogeneralagreementamongauthorsregardingwh

6、ereimage processing stops andotherrelatedareas,suchasimageanalysisandcomputervi-sion,start. Sometimesadistinctionismadebydefiningimageprocessingasadiscipline in whichboththeinputandoutputofaprocessareimages.Webelievethistobealimiting andsomewhatartificialboundary.Forexample, underthisdefinition,even

7、thetrivialtaskofcomputingtheaverageintensityofanimage(whichyieldsasinglenumber)wouldnotbeconsideredanimageprocessing operation.Ontheotherhand,therearefieldssuchascomputervision whose ultimategoalistousecomputerstoemulatehumanvision,includinglearningandbeingabletomakeinferencesandtakeactions based on

8、visualinputs.Thisareaitselfisabranchofartificialintelligence(AI)whoseobjectiveistoemulatehumanintelligence.ThefieldofAIisinitsearlieststagesofinfancyintermsof development,withprogresshavingbeenmuchslowerthanoriginallyanticipated.Theareaofimageanalysis( also called image understanding )isinbe-tweenim

9、ageprocessingandcomputervision.Therearenoclearcutboundariesinthecontinuumfromimageprocessingatoneendtocomputervisionatthe other.However,oneusefulparadigmistoconsiderthreetypesofcomputerizedprocessesinthiscontinuum:low-,mid-,and highlevelprocesses.Lowlevel processes involveprimitiveopera-tionssuchasi

10、magepreprocessingtoreducenoise,contrastenhancement,andimagesharpening.Alowlevelprocessis characterizedbythefactthatbothitsinputsandoutputsare images. Midlevelprocessingonimagesinvolvestaskssuchassegmentation(partitioninganimageintoregionsorobjects), descriptionofthoseobjectstoreducethemtoaformsuitab

11、leforcomputerprocessing,andclassification(recognition)ofindividualobjects.Amidlevelprocessischaracterizedbythefactthatitsinputsgenerallyareimages,butitsoutputsareattributesextractedfromthoseimages(e.g.,edges,contours,andtheidentityofindividualobjects).Finally,higherlevelprocessinginvolves“makingsens

12、e”ofanensembleofrecognizedobjects,asinimageanalysis,and,atthefarendofthecontinuum,performingthecognitivefunctionsnormallyassociatedwithvision.Basedontheprecedingcomments, weseethatalogicalplaceofoverlapbetweenimageprocessingandimageanalysisistheareaofrecognitionofindividualregionsorobjectsinanimage.

13、Thus,whatwecallin thisbookdigital imageprocessingencompassesprocesseswhoseinputsandoutputsareimagesand, inaddition,encompassesprocessesthatextractattributesfromimages,uptoand includingthe recognitionof individualobjects.Asasimpleillustrationtoclarifytheseconcepts,considertheareaofautomatedanalysisof

14、text.Theprocessesofacquiringanimageoftheareacontainingthetext,preprocessingthatimage,extracting(segmenting)theindividualcharacters,describingthecharactersinaformsuitableforcomputerprocessing,andrecognizingthoseindividualcharactersareinthescopeofwhatwecalldigitalimageprocessinginthisbook.Makingsense

15、ofthecontentofthepagemaybeviewedasbeinginthedomainofimageanalysisandevencomputervision,dependingonthelevelofcomplexityimpliedbythestatement“makingsense.”Aswillbecomeevidentshortly,digitalimageprocessing,aswe havedefinedit,is usedsuccessfullyinabroadrangeofareasofexceptionalsocialandeconomicvalue.The

16、areasofapplicationofdigitalimageprocessingaresovariedthatsomeformoforganizationisdesirableinattemptingtocapturethebreadthofthisfield.Oneofthesimplestwaystodevelopabasicunderstandingoftheextentof imageprocessingapplicationsistocategorizeimagesaccordingtotheirsource(e.g.,visual,Xray,andsoon).The princ

17、ipal energysourceforimagesinusetodayistheelectromagneticenergyspectrum.Otherimportantsources of energyincludeacoustic,ultrasonic,andelectronic( intheformofelectronbeamsusedinelectronmicroscopy ). Syntheticimages ,usedformodelingandvisualization,aregeneratedbycomputer.Inthissectionwediscussbrieflyhow

18、imagesaregeneratedin thesevariouscategoriesandtheareasinwhichtheyareapplied.ImagesbasedonradiationfromtheEMspectrumarethemostfamiliar,es-peciallyimagesintheXrayandvisualbandsofthespectrum.Electromagneticwavescanbeconceptualizedaspropagatingsinusoidalwavesofvaryingwavelengths,ortheycanbethoughtofasas

19、treamofmasslessparticles,eachtravelinginawavelikepatternandmovingat thespeedof light.Eachmasslessparticlecontainsacertainamount(orbundle)ofenergy. Eachbundleofenergyiscalledaphoton.Ifspectralbandsaregroupedaccordingtoenergyperphoton,weobtainthe spectrumshowninfig.below,rangingfromgammarays(highesten

20、ergy)atoneendtoradiowaves(lowestenergy)attheother. The bandsareshownshadedtoconveythefactthatbandsofthe EMspectrumarenotdistinctbutrathertransitionsmoothlyfromonetotheother.Image acquisitionisthefirstprocess. Notethatacquisitioncouldbeassimpleasbeinggivenanimagethatisalreadyindigital form.Generally,

21、theimageacquisitionstageinvolvespreprocessing,suchasscaling.Imageenhancementisamongthesimplestandmostappealingareasofdigitalimageprocessing.Basically,theideabehindenhancementtechniquesistobringoutdetailthatisobscured,orsimplytohighlightcertainfeaturesofinterestinanimage.Afamiliarexampleofenhancement

22、iswhenweincreasethecontrastofanimagebecause“itlooksbetter.”Itisimportanttokeepinmindthatenhancementisaverysubjectiveareaofimageprocessing. Imagerestorationisanareathatalsodealswith improvingtheappearanceofanimage.However,unlikeenhancement,which is subjective ,imagerestorationisobjective,inthesenseth

23、atrestorationtechniquestendtobebasedonmathematicalorprobabilisticmodelsofimagedegradation.Enhancement,ontheotherhand,isbasedonhumansubjectivepreferencesregardingwhatconstitutesa“good”enhancementresult.Color image processing isanareathathasbeengaininginimportancebecauseofthesignificantincreaseintheus

24、eofdigitalimagesovertheInternet.Itcoversanumberoffundamentalconceptsincolormodelsandbasiccolorprocessinginadigitaldomain.Colorisusedalsoinlaterchaptersasthebasisforextractingfeaturesofinterestinanimage.Waveletsarethefoundationforrepresentingimagesinvariousdegreesofresolution.Inparticular,thismateria

25、lisusedinthisbookforimagedatacompressionandforpyramidalrepresentation,inwhichimagesaresubdividedsuccessivelyintosmallerregions.Compression, as thenameimplies,dealswithtechniquesforreducingthestoragerequiredtosaveanimage, orthe bandwidthrequiredtotransmiit.Althoughstorage technologyhasimprovedsignifi

26、cantlyoverthepastdecade,thesamecannotbesaidfortransmissioncapacity.ThisistrueparticularlyinusesoftheInternet,whicharecharacterizedby significant pictorial content.Imagecompressionisfamiliar( perhapsinadvertently )tomostusersofcomputersintheformofimagefileextensions,suchasthejpgfileextensionusedinthe

27、JPEG( JointPhotographicExpertsGroup ) imagecompressionstandard.Morphologicalprocessingdealswithtoolsforextractingimagecomponentsthatareusefulintherepresentationanddescriptionofshape.Thematerialinthischapterbegins atransitionfromprocesses that outputimages to processesthatoutputimageattributes.Segmentationprocedurespartitionanimageintoitsconstituentpartsorobjects.