圖像識(shí)別外文文獻(xiàn)翻譯

譯文學(xué)院 電氣信息工程學(xué)院 專業(yè) 計(jì)算機(jī)科學(xué)與技術(shù) 學(xué)生姓名 XXX 班級(jí)學(xué)號(hào) 1045532111 指導(dǎo)教師 XXX 二零一四年六月外文文獻(xiàn)翻譯外文文獻(xiàn)翻譯PAGEPAGE15DigitalImageProcessingandEdgeDetectionTurgayCelikHasanDemire1,HuseyinOzkaramanli,MustafaUygurogluLDigitalImageProcessingInterestindigitalimageprocessingmethodsstemsfromtwoprincipalapplicantionareas:improvementofpictorialinformationforhumaninterpretation;andprocessingofimagedataforstorage,transmission,andrepresentationforau?tenuousmachineperception.Animagemaybedefinedasatwo-dimensionalfunction,Rx,y),wherexandyarespatial(planc)coordinates,andtheamplitudeoffatanypairofcoordinates(x,y)iscalledtheintensityorgrayleveloftheimageatthatpoint?Whenx,y,andtheamplitudevaluesoffarcallfinitc,discretequantities,wccalltheimageadigitalimage?Thefieldofdigitalimageprocessingreferstoprocessingdigitalimagesbymeansofadigitalcompute匚Notethatadigitalimageisconposedofafinitenumberofelements,eachofwhichhasaparticularlocationandvalue.Theseelementsarereferredtoaspictureelements,imageelements,peels,aixipixels?Pixelisthetermmostwidelyusedtodenotetheelementsofadigitalimage?Visionisthemostadvancedofoursenses,soitisnotsurprisingthatimagesplaythesinglemostimportantroleinhumanperception?However,unlikehumans,whoarelimitedlothevisualbandoftheelectromagnetic(EM)spec-trump,imagingmachinescoveralmosttheentireEMspectrunirangingfromgammatoradiowaves.Theycanoperateonimagesgeneratedbysourcesthathumansarenotaccustomedtoassociatingwithimages.Theseincludeultra-sound,electronmicroscopy,andcomputer-generated?Thus,digitalimageprocessingencompassesawideandvariedfieldofapplications?Thereisnogeneralagreementamongauthorsregardingwhereimageprocessingstopsandotherrelatedareas,suchasimageanalysisandcomputervi-son,start.Sometimesadistinctionismadebydefiningimageprocessingasadisciplineinwhichboththeinputandoutputofaprocessareimages?Webelievethistobealimitingandsomewhatartificialboundary.Forexample,uixlerthisdefinition,eventhetrivialtaskofcomputingtheaverageintensityofanimage(whichyieldsasinglenumber)wouldnotbeconsideredanimageprocessingoperationOntheotherhand^therearefieldssuchascomputervisionwhoseultimategoalistousecomputerstoemulatehumanvision,includinglearningandbeingabletomakeinferencesandtakeactionsbasedonvisualinputs?Thisareaitselfisabranchofartificialintelligence(Al)whoseobjectiveistoemulatehuman?ThefieldofAlisinitsearlieststagesofinfancyintermsofdevelopment,withprogressleavingbeenmuchslowerthanoriginallyanticipated?Theareaofimageanalysis(alsocalledimageunderstanding)isinbc?teenimageprocessingandcomputervision?Therearcnoclcar-cutboundariesinthecontinuumfromimageprocessingatoneendtocomputervisionattheother.However,oneusefulparadigmistoconsiderthreetypesofcomputerizedprocessesinthiscontinuum:low-,mid-,andhigh?levelprocesses.Low-levelprocessesinvolveprimitiveopera?tonssuchasimagepreprocessingtoreducenoise,contrastenhancement,andimagesharpening?Alow-levelprocessischaracterizedbythefeetthatbothitsinputsandoutputsareimages.Mid-levelprocessingonimagesinvolvestaskssuchassegmentation(partitioninganimageintoregionsorobjects),descriptionofthoseobjectstoreducethemtoaformsuitableforcomputerprocessing,andclassification(recognition)ofindividualobjects.Amidlevelprocessischaracterizedbythefactthatitsinputsgenerallyareimages,butitsoutputsareattributesextractedfromthoseimages(e.g.,edges,contours,andtheidentityofindividualobjects).Finally,higher-levelprocessinginvolves"makingsense"ofanensembleofrecognizedobjects,asinimageanalysis,and,attheforendofthecontinuum,performingthecognitivefunctionsnormallyassociatedwithvision.Basedontheprecedingcomments,wcseethatalogicalplaceofoverlapbetweenprocessingandimageanalysisistheareaofrecognitionofindividualregionsorobjectsinanimage?Thus,whatwecallinthisbookdigitalimageprocessingencompassesprocesseswhoseinputsand9outputsareimagesand,inadditioneneompassesprocessesthatextractattributesfromimages,up9toandincludingtherecognitionofindividualobjects.Asasimpleillustrationtoclarifytheseconcepts.considertheareaofautomatedanalysisoftext.Theprocessesofacquiringanimageoftheareacontainingthetext,preprocessingthatimage,extracting(segmenting)theindividualcharacters,describingthecharaclersinaformsuitableforcomputerprocessing,andrecognizingthoseindividualcharactersareinthescopeofwhatwecalldigitalimageprocessinginthisbook.Makingsenseofthecontentofthepagemaybeviewedasbeinginthedomainofimageanalysisandevencomputervisioadepeixiingonthelevelofcomplexityimpliedbythestatement"makingsense."Aswillbecomeevidentshortly,digitalimageprocessing,aswehavedefinedit,isusedsuccessfullyinabroadrangeofareasofexceptionalsocialandeconomicvalue.Theareasofapplicationofdigitalimageprocessingarcsovariedthatsomeformoforganizationisdesirableinattemptingtocapturethebreadthofthisfield.Oneofthesimplestwaystodevelopabasicunderstandingoftheextentofimageprocessingapplicationsistocategorizeimagesaccordingtotheirsoiucc(c.g，visual.X-ray,andsoon).TheprincipalenergysourceforimagesinusetodayistheelectromagneticenergyspectrumOtherimportantsourcesofenergyincludeacoustic,ultrasonic,andelectronic(intheformofelectronbeamsusedinelectronmicroscopy) Syntheticimages,usedformodelingandvisualization^aregeneratedbycompute匚Inthissectionwediscussbrieflyhowimagesaregeneratedinthesevariouscategoriesandtheareasinwhichtheyareapplied?ImagesbasedonradiationfromtheEMspectrumarethemostfamiliar,especiallyimagesintheX-rayandvisualbandsofthespectrumElectromagneticewavescanbeconceptualizedpropagatingsinusoidalwavesofvaryingwavelengtlis,ortheycanbethoughtofasastreamofmasslessparticles,eachtravelinginawavelikepatternandmovingatthespeedoflightEachmasslessparticlecontainsacertainamount(orbundle)ofenergy?Eachbundleofenergyiscalledaphoton.【fspectralbandsarcgroupedaccordingtoenergyperphoton,wcobtainthespectrumshowninfig.below,rangingfromgammarays(highestenergy)atoneendtoradiowaves(lowestenergy)attheother.ThebandsarcshownshadedtoconveythefactthatbandsoftheEMspectrumarcnotdistinctbutrathertransitionsmoothlyfi*omonetotheother.Imageacquisitionisthefirstprocess?NotethatacquisitioncouldbeassimpleasbeinggivenanimagethatisalreadyindigitalformGenerally,theimageacquisitionstageinvolvespreprocessing,suchasscaling?Imageenhancementisamongthesimplestandmostappealingareasofdigitalimageprocessing.Basically,theideabehindenhancementtechniquesistobringoutdetailthatisobscured,orsimplytohighlightcertainfeaturesofinterestinanimage.Afamiliarexampleofenhancementiswhenweincreasethecontrastofanimagebecause"itlooksbetter.^Itisimportanttokeepinmindthatenhancementisaverysubjectiveareaofimageprocessing.Imagerestorationisanareathatalsodealswithimprovingtheappearanceofanimage?However,unlikeenhancement,whichissubjective,imagerestorationisobjective,inthesensethatrestorationtechniquestendtobebasedonmathematicalorprobabilisticmodelsofimagedegradation.Enhanccincnt,ontheotherhand,isbasedonhumansubjectivepreferencesregardingwhatconstitutesa“good"enhanceentresultColorimageprocessingisanareathatliasbeengaininginimportaneebecauseofsignificantincreaseintheuseofdigitalimagesovertheInternet.Itcoversanumberoffuixiamentalconceptsincolormodelsandbasiccolorprocessinginadigitaldomain?Colorisusedalsoinlaterchaptersasthebasislorextractingfeaturesofinterestinanimage.Waveletsarethefoundationforrepresentingimagesinvariousdegreesofresolution.Inparticular,thismaterialisusedinthisbookforimagedatacompressionandforpyramidalrepresentation,inwhichimagesaresubdividedsuccessivelyintosmallerregions.Compression,asthenameimplies,dealswithtechniquesforreducingthestoragerequiredsavinganimage,orlhebandwidthrequiredtransmittingit.Althoughstoragetechnologyhasimprovedsignificantlyoverthepastdecade,thesamecannotbesaidfortransmissioncapacity?ThisistrueparticularlyinusesoftheInternet,whicharecharacterizedbysignificantpictorialcontent.Imagecompressionisfemiliar(perhapsinadvertently)tomostusersofcomputersintheformofimagefileextensions,suchasthejpgfileextensionusedintheJPEG(JointPhotographicExpertsGroup)imagecomprcssionstandard?Moipliologicalprocessingdealswithtoolsforextractingimagecomponentsthatarcusefulintherepresentationanddescriptionofsliape.Thematerialinthischapterbeginsatransitionfromprocessesthatoutputimagestoprocessesthatoutputimageattributes.Segmentationprocedurespartitionanimageintoitsconstituentpartsorobjects.Ingeneral,autonomoussegmentationisoneofthemostdifficulttasksindigitalimageprocessing.Aruggedsegmentationprocedurebringstheprocessalongwaytowardsuccessfulsolutionofimagingproblemsthatrequireobjectstobeidentifiedindividually.Ontheotherhand,weakorerraticsegmentationalgorithmsalmostalwaysguaranteeeventual?Ingeneral,themoreaccuratethesegmentation,themorelikelyrecognitionistosucceed?Representationanddescriptionalmostahvaysfollowtheoutputofasegmentationstage,whichusuallyisrawpixeldata,constitutingeitherthebound-rayofaregion(i.e.,thesetofpixelsseparatingoneimageregionfi*omanother)orallthepointsintheregionitselfIneithercase,convertingthedatatoaformsuitableforconiputcrprocessingisnecessary.Thefirstdecisionthatmustbemadeiswhetherthedatashouldberepresentedasaboundaryorasacompleteregion.Boundaryrepresentationisappropriatewhenthefocusisonexternalshapecharacteristics,suchascornersandinflections ?Regionalrepresentationisappropriatewhenthefocusisoninternalproperties,suchastextureorskeletalshape?Insomeapplications,theserepresentationscomplementeachother.Choosingarepresentationisonlypartofthesolutionfortrans-formingrawdataintoaformsuitableforsubsequentcomputerprocessing.Amethodmustalsobespecifiedfordescribingthedatasothatfeaturesofinterestare?Description,alsocalledfeatureselection,dealswithextractingattributesthatresultinsomequantitativeinformationofinterestorarebasicfordifferentiatingoneclassofobjectsfromanother.Recognitionistheprocessthatassignsalabel(e.g?,"vehicle')toanobjectbasedonitsdescriptors.Asdetailedbefore,weconcludeourcoverageofdigitalimageprocessingwiththedevelopmentofmethodsforrecognitionofindividualobjects?SofarwchavesaidnothingabouttheneedforpriorknowledgeorabouttheinteractionbchvccntheknowledgebaseandtheprocessingmodulesinFig2

Knowledgeaboutaproblemdomainiscodedintoanimageprocessingsystemintheidrmofaknowledgedatabase?Thisknowledgemaybeasslim-pleaasdetailingregionsofanimagewheretheinformationofinterestisknowntobelocated,thuslimitingthesearchthathastobeconductedinseekingthatinformiition.Theknowledgebasealsocanbequitecomplex,suchasaninterrelatedlistofallmajorpossibledefectsinamaterialsinspectionproblemoranimagedatabasecontaininghigh--resolutionsatelliteimagesofaregionincon?lectionwithchange-detectionapplications?Inadditiontoguidingtheoperationofeachprocessingmodule,theknowledgebasealsocontrolstheinteractionbehveenmodules.ThisdistinctionismadeinFig2abovebytheuseofdouble-headedarrowsbetweentheprocessingmodulesandtheknowledgebase,asop-posedtosingle-headedarrowslinkingtheprocessingmodules.2.EdgedetectionTheimageedgeisoneofimagemostbasiccharacteristics,ofteniscarryingimagemajorityofinfbrmationsoButtheedgeexistsintheimageirregularstructureandinnotthesteadyphenomenon,alsonamelyexistsinthesignalpointofdiscontinuityplace,thesespotshavegiventheimageoutlincposition,theseoutlinesarcfrequentlywewhentheimageryprocessingneedstheextremelyimportantsomerepresentativecondition,thisneedsustoexamineandtowithdrawitsedgetoanimage 。Buttheedgeexaminationalgorithmisintheimageryprocessingquestiononeofchssicaltechnicaldiflicultproblems,itssolutioncaniesonthehighlevelregardingusthecharacteristicdescription,therecognitionandtheunderstandingandsoonhasthesignificantinfluence;Alsobecausetheedgeexaminationallhasinmanyaspectstheextremelyimportantusevalue,thereforehowthepeoplearedevotingcontinuouslyinstudyandsolvelhestructuretoleavehavethegoodnatureandthegoodeilectedgeexaminationoperator。Intheusualsituation,wemaythesignalinsingularpointandthepointofdiscontinuitythoughtisintheimageperipheralpoint,itsnearbygradationchangesituationmayreflectfromitsneighboringpictureelementgradationdistributiongradientoEdgedctcctionisaterminologyinimageprocessingandcomputervision,particularlyintheareasoffeaturedetectionandfeatureextraction,torefertoalgorithmswhichaimatidentifyingpointsinadigitalimageatwhichtheimagebrightnesscliangcssharplyormoreformallyhasdiscontinuities.Althoughpointandlinedctcctioncertainlyarcinportantinanydiscussiononsegmentation,detectionisbyfarthemostcommonapproachfordetectingmeaningfuldiscountiesingraylevel.Theimagemajoritymaininformational!existsintheimageedge,themainperformancefortheimagepartialctoacteristicdiscontinuityisintheimagethegradationcliangequitefierceplace,alsoisthesignalwhichweusuallysaidhasthestrangechangeplace。Thestrangesignalthegradationchangewhichmovestowardsalongtheedgeisfierce,usuallywedividetheedgeforthestepsliapeandtheroofshapetwokindoftypes(asshowninFigurel-l).Inthestepedgetwosidegreylevelshavetheobviouschange;Buttheroofshapeedgeislocatedthegradationincreaseandthereducedintersectionpoint.Mayportraytheperipheralpointinmathematicsusingthegradationderivativethecliange,tothestepedge,theroofshapeedgeasksitsstep,thesecondtimederivative。Toanedge,hasthepossibilitysimultaneouslytohavethestepandthelineedgecharacteristicForexampleonasurface,cliangcsfromaplanetothenormaldirectiondifferentanotherplanecanproducethestepedge;Ifthissurlacchastheedgesandcornerswhichtheregularreflectioncharacteristicalsotwoplanesformquitetobesmooth,thenworksaswhenedgesandcornerssmoothsurfacenormalafterminorsurfacereflectionangle,asaresultoftheregularreflectioncomponent,canproducethebrightlightstripontheedgesaiuicornerssmoothsurface,suchedgelookedlikehaslikelysuperimposedalineedgeinthestepedge Becauseedgepossibleandinsceneobjectimportantcharacteristiccorrespondence，thereforeitistheveryimportantimagecharacteristicoForinstance,anobjectoutlineusuallyproducesthestepedge,becausetheobjectimageintensityisdifferentwiththebackgrouixiimageintensityoWeknewthat,theedgeexaminationessenceisusessomealgorithmtowithdrawintheimagetheobjectandthebackgroundjunctiondemarcationline.Wedefinetheedgefortheimageinthegradationoccurtherapidchangeregionboundary.Theimagegradationchangesituationmayuse什imagegradationdistributionthegradienttoreflect,thereforewemayusethepartialimagedifferentialtechnologytoobtaintheedgeexaminationopcrator.Theedgeexaminationalgorithmhasthefollowingfoursteps:Filter:Theedgeexaminationalgorithmmainlyisbasedonanimageintensitystepandthesecondtimederivative,butthederivativecomputationisverysensitivetothenoise,therefbremustusethefiltertoinproveandthenoiserelatededgedetectorperfbrnianee.Needstopointoutthat,themajorityfilterhaveakocausedtheedgeintensitylosswhileiK)isereduction,therefore,strengthenstheedgeandbetweenthenoisereductionneedscompromised.Enhancement:Strengthenstheedgethefoundationisdeterminestheimageeachneighborhoodintensitythechangevalue.Theenhancementalgorithmmay(orpartkil)theintensityvaluehastheneighborhoodtheremarkablechangespottorevealsuddenly.Theedgestrengthensisgenerallycompletesthroughthecomputationgradientpeak-to-peakvalue.Pl-1ProcessingresultExamination:Hasmanypointgradientpeak-tpeakvalueintheimagequitetobebigbuttheseinthespecificapplicationdomainnotallistheedge,theretoreshouldusesomemethodtodeterminewhichselectistheperipheralpoints.Thesimpleedgeexaminationcriterionisthegradientpcak-to-pcakvaluethresholdvaluecriterion?Localization:Ifsomeapplicationsituationrequestdefiniteedgeposition,thentheedgepositionmaycomeuptheestimateinthesub-pictureelementresolution,theedgepositionalsomayestimate.Intheedgeexaminationalgorithm,thefirstthreestepsuseextremelyuniversally.Thisisbecauseundertliemajoritysituations,needstheedgedetectortopointoutmerelytheedgeappearsinimagesomepictureelementneighbor,butisnotunnecessarytopointouttheedgetheexactlocationorthedirection.Theedgeexaminestheerrorusuallyisreferstotheedgetoclassifytheerrorbymistake,namelydistinguishedthevacationedgetheedgeretains,butdistinguishedtherealedgethevacationedgeremoves.Theedgeerrorofestimationisdescribestheedgepositionandthelateralerrorwiththeprobabilitystatisticalmodel.Weexaminetheedgetheerrorandtheedgeerrorofestimationdifferentiate,isbecausetheircomputationalmethodiscompletelydifferent,itserrormodelcompletelyisalsodifferent?Theedgeexaminationisexaminestheimagepartialremarkablechangcthemostfuixiamcntaloperation.Intheunidiincnsionalsituation,thestepedgeconcernswiththeimagefirstderivativepartialpeakvalue.Thegradientisthefiinctionchangeonekindofmeasuie,butanimagemayregardasistheimageintensitycontinuousfunctionsamplingpointarray.Therefore,issimilarwiththeunidimensionalsituation,theimagegreylevelremarkablechangeavailablegradientdiscreteapproximationiiinctionexamines?Thegradientisfirstderivativetwo-dimensionalequivalent-like,definesforthevectorasaresultofeachkindofreason,theimagealwaysreceivesthestochasticnoisethedisturbance,maysaythenoiseisubiquitous.Becausetheclassicaledgeexaminationmethodhasintroducedeachformdifferentiate,thuscausesinevitablytothenoiseextremelyscrisitivc,carriesouttheedgeexaminationresultisfrequentlyexaminesthenoiseregardperipheralpoint,butbutthegenuineedgealsoasaresultofreceivesthenoisejammingnottoexamine.Thusregardinghasthenoiseimage,onegoodedgeexaminationmethodshouldhavethegoodnoiseabatementability,simultaneouslyalsohasthecompleteedgemaintenancecharacteristic。Accordingtothischaracteristic,weproposedmanykindsofedgeexaminationoperator:IfRobertoperator,Sobeloperator,Prewittoperator,Laplaceoperatorandsoon.ThesemethodsmanyarewaitfortheprocessingpictureelementtocaiTyonthegradationanalysisforthecentralneighborhoodachievementthefoundation,realizedandhasalreadyobtainedthegoodprocessingefleettotheimageedgeextraction,oButthiskindofmethodsimultaneouslyalsoexistshastheedgepictureelementwidth,thenoisejammingisseriousandsoontheshortcomings,evenifusessomeauxiliarymethodstoperformthedenoising,alsocorrespondingcanbringtheflawwhichtheedgefuzzyandsoonovercomeswithdifficulty。Alongwiththewaveletanalysisappearancetitsgoodtimefrequencypartialcliaracteristicbythewidespreadapplicationintheimageryprocessingandinthepatternrecognitiondomain,becomesinthesignalprocessingthecommonlyusedmethodandthepowerfultoolThroughthewaveletanalysis,mayinterweavedecomposesinthesameplaceeachkindofcompositesignalthedifferentfrequencytheblocksignal,butcarriesontheedgeexaminationthroughthewavelettransformation,mayuseitsmulti-criteriaandthemulti-resolutionnatureftilly,realeffectiveexpressestheimagetheedgecharacteristicoWhenthewavelettransformationcriterionreduces,ismoresensitivetotheimagedetail;Butwhenthecriterionincreases,theimagedetailisfilteredout,theexaminationedgewillbeonlythethickoutline.Thischaracteristicisextremelyusefulinthepatternrecognition,wemaybecalledthisthickoutlinetheimagethemainedge.Ifwillbeableanimagemainedgeclearintegrityextractioiithistothegoaldivision,therecognitionandsoonfollowingprocessingtobringtheenormousconvenience.Generallyspeaking,theabovemethodallistheworkwhichdoesbasedontheimageluminanceinlormationoInthemultitudinousscientiticresearchworkerunder,hasobtainedtheverygoodeffectdiligently.But,becausetheimageedgereceivesphysicalconditionandsoontheilluminationinfluencesquitetobebigabove,oftenenablesmanytohaveacommonshortcomingbasedonbrightnessedgedetectionmethod,tliatistheedgeisnotcontinual,docsnotsealup.Considcrcdthephaseinformationintheimageimportanceaswellasitsstablecharacteristic,causesusingthephaseinformationtocarryontheimageryprocessingintonewresearchtopic。Inthispapersoonintroducesonekindbasedonthephaseimagecliaracteristicexaminationmethodphaseuniformniethod.ltisnotusestheimagetheluminanceinformation,butisitspliasecharacteristic,namelysuppositionimageFouriercomponentphasemostconsistentspotachievementcharacteristicpoint.Notonlyitcanexaminebrightnesscharacteristicsandsoonstepcharacteristic,linecharacteristic,lwreovercanexamineMachbeltphenomenonwhichproducesasaresultofthehumanvisionsensationcharacteristic.Becausethephaseuniiormitydoesnotneedtocarryonanysuppositiontotheimagecharacteristictype,thereforeithastheverystrong。Doesnothaveinthenoisesituationintheimage,thePrewittoperator,theRobertoperator,theSobeloperatoraswellasthedifferentialgradientoperator,allcanthequiteaccurateexaminationedge.But,afterjoinsthewhitegaussiannoise,theRobertoperatorreceivestheinfluciKCissmallest,nextisthePrewittoperator,receivesaffectsinabigwayistheSobeloperator,butregardingthedifferentialgradientoperator,thenistheimageoverallcontrastgradienthasobviousdcprcssion.2-17mayseebyFigurc,indocsnothaveinthenoisesituation,theCannyoperator,theLOGoperatorandtheLaplaceoperatorallmayobtainthequitegoodexaminationeffect,but,theLOGoperatoraKvayscanproducethefalseedge,thisanditszerocrossingexaminationmethodconcerns.Afteraddsonthenoise,traditionalexiiminationoperator(Laplaceoperator)theexaminationqualitydroppedobviously,buttheLOGoperatorhasproducedmorefalseedgesunderthenoiseconditionButtheCannyoperatorexaminationresultiscontinuouslyextremelysatisfying?Aboutthenoisetotheedgealgorithminfluence,weinthenextsection,taketheLOGalgorithmastheexample,makesthequiteexhaustiveanalysisandtheelaboration.Theimagemajoritymaininformationallexistsintheimageedge,themainperformancefortheimagepartialcharacteristicdiscontinuity;isintheimagethegradationchangequitefierceplace,alsoisthesignalwhichweusuallysaidhasthestrangechangeplaceoThestrangesignalthegradationchangewhichmovestowardsalongtheedgeisfierce,usuallywedividetheedgeforthestepshapeandtheroofshapetwokindoftypes(asshowninFigurc1?1).Inthestepedgetwosidegreylevelshavetheobviouschange;Buttheroofshapeedgeislocatedthegradationincreaseandthereducedintersectionpoint.Mayportraytheperipheralpointinmathematicsusingthegradationderivativethechange,tothestepedge,theroofshapeedgeasksitsstep,thesecondtimederivativeseparatelyaToanedge,hasthepossibilitysimultaneouslytohavethestepandthelineedgecharacteristic.Forexampleonasurface,cliangesfromaplanetothenormaldirectiondifitirentanotherplanecanproducethestepedge;Ifthissurfacehastheedgesandcornerswhichtheregularreflectioncharacteristicalsotwoplanesformquitetobesmooth,thenworksaswhenedgesandcornerssmoothsuriacenormalaftermirrorsurfacereflectionangle,asaresultoftheregularreflectioncomponent,canproducethebrightlightstripontheedgesandcornerssmoothsurfiice,suchedgelookedlikehaslikelysuperimposedalineedgeinthestepedge.Becauseedgepossibleandinsceneobjectimportantcharacteristiccorrespondence,thereforeitistheveryimportantimagecharacteristic。Forinstance,anobjectoutlineusuallyproducesthestepedge,becausetheobjectimageintensityisdifferentwiththebackgroutxlimageintensity。Althoughtheedgedetectionhadthedifferentialgradientoperator,theLaplaceoperator,theSobeloperator,theLOGoperatoraswellastheCannyoperatorandsoonmanymethods,butthesealgorithmsdonothavetheautomaticfocalvariationthought?Bu(infact,asaresultofreasonsandsoonphysicsandilluminationineachimageedgeusuallyproducesinthedifferentcriterionscope,formsthedifferenttypetheedge(forexampleedgesandsoonstep,roof),theseinformationsareunknown.Moreover,intheimagealwayshasthenoise,therefore,accordingtotheimagecharacteristic,canauto-adaptedexaminetheimagetheedgeiscorrectlyextremelydifficult ?Easytoimagine,isnotimpossibletoexaminealledgeswi