Definition
Computervisionisasimulationofbiologicalvisionusingcomputersandrelatedequipment.Itsmaintaskistoobtainthree-dimensionalinformationofthecorrespondingscenebyprocessingthecollectedpicturesorvideos,justlikehumansandmanyotherkindsofcreaturesdoeveryday.
Computervisionisasubjectofhowtousecamerasandcomputerstoobtainthedataandinformationweneedaboutthesubject.Toputitvividly,itistoinstalleyes(cameras)andbrains(algorithms)onthecomputersothatthecomputercanperceivetheenvironment.TheChineseidiom"seeingisbelieving"andthewesternsaying"Onepictureisworthtenthousandwords"expresstheimportanceofvisiontomankind.Itisnotdifficulttoimaginehowbroadtheapplicationprospectsofmachineswithvisioncanbe.
Computervisionisnotonlyanengineeringfield,butalsoachallengingandimportantresearchfieldinthescientificfield.Computervisionisacomprehensivediscipline,ithasattractedresearchersfromvariousdisciplinestoparticipateinitsresearch.Theseincludecomputerscienceandengineering,signalprocessing,physics,appliedmathematicsandstatistics,neurophysiologyandcognitivescience.
Analysis
Visionisanintegralpartofvariousintelligent/autonomoussystemsinvariousapplicationfields,suchasmanufacturing,inspection,documentanalysis,medicaldiagnosis,andmilitary.Becauseofitsimportance,someadvancedcountries,suchastheUnitedStates,listcomputervisionresearchasamajorbasicprobleminscienceandengineeringthathasawide-rangingimpactoneconomyandscience,theso-calledgrandchallenge.Thechallengeofcomputervisionistodevelopvisioncapabilitiescomparabletohumansforcomputersandrobots.Machinevisionrequiresimagesignals,textureandcolormodeling,geometricprocessingandreasoning,andobjectmodeling.Acapablevisionsystemshouldtightlyintegratealltheseprocesses.Asadiscipline,computervisionbeganintheearly1960s,butmanyimportantadvancesinthebasicresearchofcomputervisionweremadeinthe1980s.Computervisioniscloselyrelatedtohumanvision.Acorrectunderstandingofhumanvisionwillbeverybeneficialtotheresearchofcomputervision.Forthiswewillfirstintroducehumanvision.
Principium
Computervisionistheuseofvariousimagingsystemsinsteadofvisualorgansasinput-sensitivemeans,andcomputersinsteadofthebraintocompleteprocessingandinterpretation.Theultimateresearchgoalofcomputervisionistoenablecomputerstoobserveandunderstandtheworldthroughvisionlikehumans,andhavetheabilitytoadapttotheenvironmentautonomously.Agoalthatcanonlybeachievedafterlong-termefforts.Therefore,beforeachievingthefinalgoal,themid-termgoalofpeople'seffortsistoestablishavisionsystemthatcancompletecertaintasksbasedonacertaindegreeofintelligencewithvisualsensitivityandfeedback.Forexample,animportantapplicationareaofcomputervisionisthevisualnavigationofautonomousvehicles.Thereisnoconditiontorealizeasystemthatcanrecognizeandunderstandanyenvironmentandcompleteautonomousnavigationlikehumans.Therefore,theresearchgoalofpeople'seffortsistoachieveavisualassisteddrivingsystemthathasroadtrackingcapabilitiesonexpresswaysandcanavoidcollisionswithvehiclesinfront.Thepointtobepointedouthereisthatinthecomputervisionsystem,thecomputerreplacesthehumanbrain,butitdoesnotmeanthatthecomputermustcompletetheprocessingofvisualinformationaccordingtothemethodofhumanvision.Computervisioncanandshouldprocessvisualinformationaccordingtothecharacteristicsofthecomputersystem.However,thehumanvisualsystemisbyfarthemostpowerfulandcompletevisualsystemknowntopeople.Asyouwillseeinthefollowingchapters,thestudyofhumanvisualprocessingmechanismswillprovideinspirationandguidanceforcomputervisionresearch.Therefore,thecomputerinformationprocessingmethodisusedtostudythemechanismofhumanvisionandestablishthecalculationtheoryofhumanvision.ResearchinthisareaiscalledComputationalVision.Computationalvisioncanbeconsideredasaresearchfieldincomputervision.
Related
Therearemanydisciplineswhoseresearchgoalsaresimilartoorrelatedtocomputervision.Thesesubjectsincludeimageprocessing,patternrecognitionorimagerecognition,sceneanalysis,imageunderstanding,etc.Computervisionincludesimageprocessingandpatternrecognition.Inaddition,italsoincludesthedescriptionofspatialshapes,geometricmodeling,andtheprocessofrecognition.Realizingimageunderstandingistheultimategoalofcomputervision.
Imageprocessing
Imageprocessingtechnologyconvertstheinputimageintoanotherimagewithdesiredcharacteristics.Forexample,theoutputimagecanbeprocessedtohaveahighersignal-to-noiseratio,orenhancedprocessingcanbeusedtohighlightthedetailsoftheimagetofacilitateinspectionbytheoperator.Incomputervisionresearch,imageprocessingtechnologyisoftenusedforpreprocessingandfeatureextraction.
Patternrecognition
Patternrecognitiontechnologydividesimagesintopredeterminedcategoriesbasedonthestatisticalcharacteristicsorstructuralinformationextractedfromtheimage.Forexample,textrecognitionorfingerprintrecognition.Incomputervision,patternrecognitiontechnologyisoftenusedtoidentifyandclassifycertainpartsofanimage,suchassegmentedregions.
Imageunderstanding
Givenanimage,theimageunderstandingprogramnotonlydescribestheimageitself,butalsodescribesandinterpretsthescenerepresentedbytheimage,inordertomakeananalysisofthecontentrepresentedbytheimage.Decide.Intheearlydaysofartificialintelligencevisionresearch,thetermsceneanalysiswasoftenusedtoemphasizethedifferencebetweentwo-dimensionalimagesandthree-dimensionalscenes.Inadditiontocompleximageprocessing,imageunderstandingalsorequiresknowledgeaboutthephysicallawsofsceneimagingandknowledgerelatedtothecontentofthescene.
Whenestablishingacomputervisionsystem,itisnecessarytousetherelevanttechnologiesintheabovedisciplines,butthecontentofcomputervisionresearchismoreextensivethanthesedisciplines.Theresearchofcomputervisioniscloselyrelatedtotheresearchofhumanvision.Inordertoachievethegoalofestablishingageneral-purposecomputervisionsystemsimilartothehumanvisionsystem,itisnecessarytoestablishacomputertheoryofhumanvision.
Nunc et situ
Theoutstandingfeatureofthecomputervisionfieldisitsdiversityandimperfection.Pioneersinthisfieldcanbetracedbacktoearliertimes,butitwasnotuntilthelate1970swhentheperformanceofcomputerswasimprovedtohandlelarge-scaledatasuchasimagesthatcomputervisionreceivedformalattentionanddevelopment.However,thesedevelopmentsoftenoriginatefromtheneedsofotherdifferentfields,sowhatismeantby"computervisionproblems"hasneverbeenformallydefined.Naturally,thereisnoformulaforhow"computervisionproblems"shouldbesolved.
Nevertheless,peoplehavebeguntomastersomeofthemethodstosolvespecificcomputervisiontasks.Unfortunately,thesemethodsareusuallyonlyapplicabletoagroupofnarrowtargets(suchas:faces,fingerprints,text,etc.),sotheycannotbeWidelyusedindifferentoccasions.
Theapplicationofthesemethodsisusuallyacomponentofsomelarge-scalesystemsthatsolvecomplexproblems(suchasmedicalimageprocessing,qualitycontrolandmeasurementinindustrialmanufacturing).Inmostpracticalapplicationsofcomputervision,computersarepresettosolvespecifictasks.However,methodsbasedonmachinelearningarebecomingmoreandmorepopular.Oncetheresearchofmachinelearningisfurtherdeveloped,thefuture"generalpurpose"computervisionapplicationsmaybeabletocometrue.
Oneofthemainissuesstudiedbyartificialintelligenceis:howtomakethesystemhave"planning"and"decision-makingcapabilities"?Soastomakeitcompleteaspecifictechnicalaction(forexample:movearobotthroughaspecificenvironment).Thisproblemiscloselyrelatedtothecomputervisionproblem.Here,thecomputervisionsystemactsasaperceptron,providinginformationfordecision-making.Otherresearchdirectionsincludepatternrecognitionandmachinelearning(whichalsobelongtothefieldofartificialintelligence,buthaveanimportantconnectionwithcomputervision).Asaresult,computervisionisoftenregardedasabranchofartificialintelligenceandcomputerscience.
Physicsisanotherfieldthathasanimportantconnectionwithcomputervision.
Thegoalofcomputervisionistofullyunderstandtheelectromagneticwaves-mainlyvisiblelightandinfraredlight-theimageformedbythereflectionofthesurfaceoftheobject,andthisprocessisbasedonopticalphysicsandsolid-statephysics.Somecutting-edgeimageperceptionsystemswillevenbeappliedtoquantummechanicstheorytoanalyzetherealworldrepresentedbyimages.Atthesametime,manymeasurementproblemsinphysicscanalsobesolvedbycomputervision,suchasfluidmotion.Becauseofthis,computervisioncanalsobeseenasanextensionofphysics.
Anotherimportantfieldisneurobiology,especiallythepartofthebiologicalvisualsystem.
Throughoutthe20thcentury,humanshaveconductedextensivestudiesontheeyes,neurons,andbraintissuesofvariousanimalsrelatedtovisualstimulation.Thesestudieshaveledtosome"natural"Thedescriptionofhowthevisualsystemworks(althoughitisstillabitrough)hasalsoformedasub-fieldofcomputervision-peopletrytobuildartificialsystemsthatcansimulatethevisualoperationsoflivingbeingswithvaryingdegreesofcomplexity.Atthesametime,inthefieldofcomputervision,somemethodsbasedonmachinelearningalsorefertosomebiologicalmechanisms.
Anotherrelatedfieldofcomputervisionissignalprocessing.Manyprocessingmethodsrelatedtounitvariablesignals,especiallytheprocessingoftime-varyingsignals,cannaturallybeextendedtotheprocessingmethodsofbinaryvariablesignalsormultivariatesignalsincomputervision.However,duetotheuniquepropertiesofimagedata,manymethodsdevelopedincomputervisioncannotfindacorrespondingversionintheunitsignalprocessingmethod.Oneofthemaincharacteristicsofthesemethodsistheirnon-linearityandthemulti-dimensionalityofimageinformation.Theabovetwopoints,aspartofcomputervision,formaspecialresearchdirectioninsignalprocessing.
Inadditiontothefieldsmentionedabove,manyresearchtopicscanalsobetreatedaspurelymathematicalproblems.Forexample,manyproblemsincomputervisionarebasedonstatistics,optimizationtheory,andgeometry.
Howtoimplementexistingmethodsthroughvarioussoftwareandhardware,orhowtomodifythesemethods,soastoobtainreasonableexecutionspeedwithoutlosingsufficientaccuracy,isthemainissueinthefieldofcomputervisiontoday.Subject.
Applicationem
Mankindisenteringtheinformationage,andcomputerswillincreasinglyenteralmostallfields.Ontheonehand,morepeoplewithoutprofessionalcomputertrainingalsoneedtousecomputers.Ontheotherhand,thefunctionsofcomputersaregettingstrongerandstronger,andthemethodsofusingthemaregettingmoreandmorecomplicated.Thiscreatesasharpcontradictionbetweentheflexibilityofpeopleinconversationandcommunicationandthestrictnessandrigidityrequiredwhenusingcomputers.Humanscanexchangeinformationwiththeoutsideworldthroughvision,hearing,andlanguage,andcanexpressthesamemeaningindifferentways.However,computersarerequiredtowriteprogramsstrictlyinaccordancewithvariousprogramminglanguages,sothatcomputerscanrun.Inordertoenablemorepeopletousecomplexcomputers,itisnecessarytochangethepastsituationwherepeopleadapttocomputersandmemorizecomputerusagerulesbyrote.Instead,letthecomputeradapttopeople'shabitsandrequirements,andexchangeinformationwithpeopleinthewaypeopleareusedto,thatis,letthecomputerhavetheabilitytosee,hear,andspeak.Atthistime,thecomputermusthavetheabilityoflogicalreasoninganddecision-making.Acomputerwiththeabovecapabilitiesisanintelligentcomputer.
Intelligentcomputersnotonlymakecomputersmoreconvenientforpeopletouse,butatthesametime,ifsuchcomputersareusedtocontrolvariousautomationdevices,especiallyintelligentrobots,theseautomationsystemsandintelligentrobotscanadapttotheenvironment,andTheabilitytomakedecisionsindependently.Thiscanreplacepeople'sheavyworkonvariousoccasions,orreplacepeopletocompletetasksinvariousdangerousandharshenvironments.
Applicationemsrangefromtasks,suchasindustrialmachinevisionsystems,forexample,inspectionofbottlesontheproductionlinetoacceleratethrough,researchintoartificialintelligenceandcomputersorrobots,whichcanunderstandtheworldaroundthem.Thereisasignificantoverlapinthefieldsofcomputervisionandmachinevision.Computervisioninvolvesthecoretechnologyusedinautomatedimageanalysisinmanyfields.Machinevisionusuallyreferstoaprocessthatcombinesautomaticimageanalysiswithothermethodsandtechnologiestoprovideautomaticdetectionandrobotguidanceinindustrialapplications.Inmanycomputervisionapplications,computersarepre-programmedtosolvespecifictasks,butlearning-basedmethodsarenowbecomingmoreandmorecommon.Examplesofcomputervisionapplicationsincludethoseusedinsystems:
(1) Continere processus, suchasanindustrialrobot;
(2) Navigatio, sicut carrus autonomus aut mobilerobots;
(3) Events detecta, such asveillance and people counting;
(4) Organisinginformation, forexample, indexdata basibus for images and images sequences;
(5)Modelingobjectsorenvironments,suchasmedicalimageanalysissystemsorterrainmodels;
(6) Commercium, exempli causa, cum inputandae vice-computationis humanitatis negotium est;
(7) Automaticdetection, for example, inmanufaciens applicationes.
Themostprominentapplicationareasaremedicalcomputervisionandmedicalimageprocessing.Thefeatureinformationofthisareaisextractedfromtheimagedataforthepurposeofmedicaldiagnosisofthepatient.Usually,theimagedataisintheformofmicroscopeimages,X-rayimages,angiographyimages,ultrasoundimagesandtomographicimages.Anexampleoftheinformationthatcanbeextractedfromsuchimagedataisthedetectionoftumors,atherosclerosisorothermalignantchanges.Itcanalsobethesizeoftheorgan,bloodflow,etc.Thisfieldofapplicationalsosupportsthemeasurementofmedicalresearchbyprovidingnewinformation,forexample,onthestructureofthebrain,oraboutthequalityofmedicaltreatment.Theapplicationofcomputervisioninthemedicalfieldalsoincludesenhancingimagesthatareinterpretedbyhumans,suchasultrasoundimagesorX-rayimages,toreducetheeffectsofnoise.
Thesecondapplicationareaofcomputervisionisinindustry,sometimescalledmachinevision,whereinformationisextractedtosupportthepurposeofthemanufacturingprocess.Anexampleisqualitycontrol,wheretheinformationorfinalproductisautomaticallydetectedinordertofinddefects.Anotherexampleisthatthepositionanddetailorientationbeingpickeduparemeasuredbytheroboticarm.Machinevisionisalsousedextensivelyintheprocessofagriculture,frombulkmaterials,thisprocessiscalledtheremovalofunwantedthings,opticalsortingoffood.
Militaryapplicationsareprobablyoneofthelargestareasofcomputervision.Themostobviousexampleisthedetectionofenemysoldiersorvehiclesandmissileguidance.Moreadvancedsystemsguidethemissiletotheareawherethemissileissent,ratherthanaspecifictarget,andmakeaselectionwhenthemissilereachesthetargetintheareabasedonlocallyacquiredimagedata.Modernmilitaryconcepts,suchas"battlefieldperception",meanthatvarioussensors,includingimagesensors,provideawealthofrelevantcombatscenariosthatcanbeusedtosupportstrategicdecision-makinginformation.Inthiscase,automaticdataprocessingisusedtoreducecomplexityandfuseinformationfrommultiplesensorstoimprovereliability.
Anewerapplicationareaisautonomousvehicles,whichincludediving,landvehicles(smallrobotswithwheels,carsortrucks),aerialworkvehiclesandunmannedaerialvehicles(UAV).Thelevelofautonomyrangesfromcompletelyindependent(unmanned)vehiclestocars,wherecomputervision-basedsystemssupportdriverprogramsorexperimentsindifferentsituations.Afullyautonomouscarusuallyusescomputervisiontonavigatewhenitknowswhereitis,ortheenvironmentusedforproduction(mapSLAM)andfordetectingobstacles.Itcanalsobeusedtodetectspecificeventsforspecifictasks,forexample,aUAVlookingforforestfires.Examplesofsupportsystemsarecarsinobstaclewarningsystems,andautonomouslandingsystemsforaircraft.Severalautomakershavedemonstratedthesystem'sautonomousdrivingofcars,butthetechnologyhasnotreachedacertainlevelbeforeitcanbeputonthemarket.Thereareplentyofexamplesofmilitaryautonomousmodels,fromadvancedmissiles,unmannedaerialvehiclesforreconnaissancemissionsormissileguidance.Spaceexplorationisalreadyusingcomputervision,autonomousvehiclessuchasNASA’sMarsExplorationRoverandtheEuropeanSpaceAgency’sExoMarsMarsRover.
Otherapplicationareasinclude:
(1) Movies and broadcasts that support the production of visualeffects, forexemple,cameratracking(motionmatching).
(2) Cras.
Similitudines anddifferences
Computervision,imageprocessing,imageanalysis,robotvisionandmachinevisionarecloselyrelateddisciplines.Ifyouopenthetextbookswiththeabovenames,youwillfindthattheyhaveaconsiderableoverlapintechnologyandapplicationareas.Thisshowsthatthebasictheoriesofthesedisciplinesareroughlythesame,anditevenmakespeoplesuspectthattheyarethesamedisciplineswithdifferentnames.
However,variousresearchinstitutions,academicjournals,conferences,andcompaniesoftenclassifythemselvesasaparticularfield,soavarietyofcharacteristicsthatdistinguishthesedisciplineshavebeenbroughtup.Amethodofdistinctionwillbegivenbelow,althoughitcannotbesaidthatthismethodofdistinctioniscompletelyaccurate.
Theresearchobjectofcomputervisionismainlyathree-dimensionalscenemappedtoasingleormultipleimages,suchasthereconstructionofathree-dimensionalscene.Theresearchofcomputervisionislargelyfocusedonthecontentoftheimage.
Theresearchobjectsofimageprocessingandimageanalysisaremainlytwo-dimensionalimages,whichrealizeimagetransformation,especiallyforpixel-leveloperations,suchasimagecontrastimprovement,edgeextraction,denoisingandgeometrictransformationssuchasimagerotation.Thisfeatureshowsthattheresearchcontentofimageprocessingorimageanalysishasnothingtodowiththespecificcontentoftheimage.
Machinevisionmainlyreferstothevisualresearchintheindustrialfield,suchasthevisionofautonomousrobots,andthevisionforinspectionandmeasurement.Thisshowsthatinthisfield,throughsoftwareandhardware,imageperceptionandcontroltheoryisoftencloselycombinedwithimageprocessingtoachieveefficientrobotcontrolorvariousreal-timeoperations.
Patternrecognitionusesvariousmethodstoextractinformationfromsignals,mainlyusingstatisticaltheories.Oneofthemaindirectionsinthisfieldistoextractinformationfromimagedata.
Thereisanotherfieldcalledimagingtechnology.Theinitialresearchcontentinthisfieldismainlytomakeimages,butsometimesalsoinvolvesimageanalysisandprocessing.Forexample,medicalimagingincludesalargenumberofimageanalysisinthemedicalfield.
Forallthesefields,apossibleprocessisthatyouworkinacomputervisionlaboratory,youareengagedinimageprocessingatwork,andfinallysolvetheproblemsinthefieldofmachinevision,andthenpublishyourresultsinAtthemeetingofpatternrecognition.
Problemata
Almosteveryspecificapplicationofcomputervisiontechnologymustsolveaseriesofthesameproblems.Theseclassicproblemsinclude:
Recognitio
Acomputervision,imageprocessingandmachinevisioncommonclassicproblemistodeterminewhetherasetofimagedatacontainsaspecificObject,imagefeatureormovementstate.Thisproblemcanusuallybesolvedautomaticallybyamachine,butsofar,thereisnosinglemethodthatcandetermineawiderangeofsituations:recognizeanyobjectinanyenvironment.Theexistingtechnologycanandcanonlywellsolvetherecognitionofspecifictargets,suchassimplegeometricpatternrecognition,facerecognition,printedorhandwrittendocumentrecognition,orvehiclerecognition.Andtheserecognitionsneedtohavespecifiedlighting,backgroundandtargetposturerequirementsinaspecificenvironment.
Generalrecognitionhasevolvedintoseveralslightlydifferentconceptsondifferentoccasions:
Recognitio(narrowsense):Foroneormorepre-definedorlearnedObjectsorobjectsarerecognized,andtheirtwo-dimensionalpositionorthree-dimensionalpostureisusuallyprovidedduringtherecognitionprocess.
Identification:Identifythesingleobjectitself.Forexample:therecognitionofacertainface,therecognitionofacertainfingerprint.
Monitoring:Discoverspecificsituationcontentfromimages.Forexample:thediscoveryofabnormalskillsincellsortissuesinmedicine,andthediscoveryofpassingvehiclesbytrafficmonitoringequipment.Monitoringisoftentodiscoverspecialareasintheimagethroughsimpleimageprocessing,whichprovidesastartingpointforsubsequentmorecomplexoperations.
Pluresspecificapplicationdirectionsidentified:
Content-basedimageextraction:Findallpicturescontainingspecifiedcontentinahugeimagecollection.Thespecifiedcontentcantakemanyforms,suchasaredroughlycircularpattern,orabicycle.Thesearchforthelatterkindofcontenthereisobviouslymorecomplicatedthantheformer,becausetheformerdescribesalow-levelintuitivevisualfeature,whilethelatterinvolvesanabstractconcept(orhigh-levelvisualfeature).Thatis,"bicycle",theobviouspointisthattheappearanceofthebicycleisnotfixed.
Poseevaluation:Evaluationofthepositionordirectionofanobjectrelativetothecamera.Forexample:theassessmentofthepostureandpositionoftheroboticarm.
Opticalcharacterrecognitionrecognizesanddiscriminatesprintedorhandwrittentextinanimage,andtheusualoutputistoconvertitintoaneasy-to-editdocumentform.
Motus
Themonitoringofobjectmotionbasedonsequenceimagesincludesmanytypes,suchas:
Selfmotion:monitorthethree-dimensionalrigidmotionofthecamera.
Imagetracking:Trackmovingobjects.
SceneReconstruction
Giventwoormoreimagesoravideoofascene,scenereconstructionseekstobuildacomputermodel/three-dimensionalmodelofthescene.Thesimplestcaseistogenerateasetofpointsinthree-dimensionalspace.Inmorecomplexsituations,acompletethree-dimensionalsurfacemodelwillbebuilt.
Imagerestoration
Thegoalofimagerestorationistoremovenoiseintheimage,suchasinstrumentnoise,blur,etc.
Systema
Thestructureofthecomputervisionsystemlargelydependsonitsspecificapplicationdirection.Someworkindependentlyandareusedtosolvespecificmeasurementorinspectionproblems;someappearasapartofalargecomplexsystem,suchasworkingwithmechanicalcontrolsystems,databasesystems,andman-machineinterfacedevices.Thespecificimplementationmethodofthecomputervisionsystemisalsodeterminedbyitsfunction-whetheritisfixedinadvanceorisautomaticallylearnedandadjustedduringoperation.However,therearesomefunctionsthatalmosteverycomputersystemneeds:
Imageacquisition
Adigitalimageisproducedbyoneormoreimagesensors,hereThesensorcanbeavarietyofphotosensitivecameras,includingremotesensingequipment,X-raytomography,radar,ultrasonicreceivers,andsoon.Dependingonthedifferentperceptrons,thegeneratedpicturecanbeanordinarytwo-dimensionalimage,athree-dimensionalimagegrouporanimagesequence.Thepixelvalueofthepictureoftencorrespondstotheintensityoflightinoneormorespectralbands(grayscaleorcolorimage),butitcanalsoberelatedtovariousphysicaldata,suchasthedepthandabsorbanceofsoundwaves,electromagneticwavesornuclearmagneticresonanceOrreflectivity.
Preprocessing
Beforeimplementingspecificcomputervisionmethodsontheimagetoextractcertainspecificinformation,oneorsomepreprocessingisoftenusedtomaketheimagemeettherequirementsofsubsequentmethodsRequire.Forexample:
Sub-sampling toensure the correct imagecoordinates;
Smoothdenoisingtofilteroutthedevicenoiseintroducedbythesensor;
ImprovethecontrasttoensuretherealizationRelevantinformationcanbedetected;
Adjustthescalespacetomaketheimagestructuresuitableforlocalapplications.
Featureextraction
Extractfeaturesofvariouscomplexityfromtheimage.Forexample:
Linea, edgeextraction;
Localizedfeaturepointdetectionsuchascornerdetection,spotdetection;
MorecomplexfeaturesmayberelatedtotheimageThetextureshapeormovementisrelated.
Detectionsegmentation
Intheprocessofimageprocessing,itissometimesnecessarytosegmenttheimagetoextractvaluablepartsforsubsequentprocessing,suchas
screeningFeaturepoints;
Segmentthepartofoneormorepicturesthatcontainsaspecifictarget.
Advancedprocessing
Atthispoint,thedataoftenhasasmallamount,suchasthepartoftheimagethatisconsideredtocontainthetargetobjectafterpreviousprocessing.Theprocessingatthistimeincludes:
Verifywhetherthedataobtainedmeetstheprerequisiterequirements;
Estimatespecificcoefficients,suchasthetarget’sattitudeandvolume;
sort.
Advancedprocessinghasthemeaningofunderstandingimagecontent.Itisahigh-levelprocessingincomputervision.Itismainlybasedonimagesegmentationtounderstandthesegmentedimageblocks,suchasrecognitionandotheroperations..
Requisita
Theinfluenceoflightsourcelayoutneedstobecarefullyconsidered.
Select the correctlensgroup,considering themagnification, space, size, distortion...
Choosetherightcamera(CCD), considerans thefunctionem, speciem, stabilitatem, firmitatem ...
Visualsoftwaredevelopmentneedstorelyontheaccumulationofexperience,trymoreandthinkaboutthewaytosolvetheproblem.
Theultimategoalistocontinuouslyimprovetheaccuracyofcreationandshortentheprocessingtime.
finem.
Colloquium
Top
ICCV:InternationalColloquiumonComputerVision,InternationalComputerVisionColloquium
CVPR:InternationalColloquiumonComputerVisionandPatternRecognitio,InternationalColloquiumonComputerVisionandPatternRecognitio
ECCV:EuropeanColloquiumonComputerVision,EuropeanColloquiumonComputerVision
Melior
ICIP:InternationalColloquiumonImageProcessing,InternationalColloquiumonImageProcessing
BMVC:BritishMachineVisionColloquium,BritishMachineVisionColloquium
ICPR:InternationalColloquiumonPatternRecognitio,InternationalColloquiumonPatternRecognitio
ACCV:AsianColloquiumonComputerVision,AsianColloquiumonComputerVision
Journal
Top
PAMI:IEEETransactionsonPatternAnalysisandMachineIntelligence,IEEEPatternAnalysisJournalofMachineIntelligence
IJCV:InternationalJournalonComputerVision,InternationalJournalofComputerVision
Melior
TIP:IEEETransactionsonImageProcessing,IEEEImageProcessingMagazine
CVIU:ComputerVisionandImageUnderstanding,ComputerVisionandImageUnderstanding
PR:PatternRecognitio,PatternRecognitio
PRL:PatternRecognitioLetters,PatternRecognitioExpress