车辆图像预处理和车牌定位的方法研究 中英文对照翻译Word文档格式.docx
《车辆图像预处理和车牌定位的方法研究 中英文对照翻译Word文档格式.docx》由会员分享,可在线阅读,更多相关《车辆图像预处理和车牌定位的方法研究 中英文对照翻译Word文档格式.docx(13页珍藏版)》请在冰豆网上搜索。
Medianfiltering;
Binaryzation;
Edgedetection;
Licenseplatelocation
I.INTRODUCTION
Vehiclelicenseplaterecognitionsystembasedonvehiclelicenseforthespecifictargetisdedicatedcomputervisionsystem[1].Itisoneoftheimportantresearchtopicsaboutcomputervisionandpatternrecognitiontechnologyinthefieldofintelligenttransportationapplications.Vehiclelicenseidentificationisthegeneralcomposedbythefollowingprocess:
imageacquisition,imagepreprocessing,licenseplatelocation,charactersegmentation,characterrecognition[2].Thecorrectrateofthelastprocesshasadirectimpactonthenextprocess.Sincetheoriginalimagefromtheacquisitioncardincludesthevehiclelicenseplate,thecaritself,andautomotivebackgroundimage,itisnecessarytoremovethesenon-licensedimagesinordertoextractthecorrectregionallicenseandforthefoundationofthelicenseplatecharacterrecognition.Intheactualsystem,duetonaturalchangesindayandnightillumination,vehicleownmovement,thecameraangleofobservation,collectingimagesoftheequipmentitselfandotherfactorsinfluence,theimageobtainedisnotalwaysverysatisfactory,thereisawiderangeofnoise.Therefore,itisnecessarytomakethelicenseplateimagepre-processingforimprovingimagequality,layingfoundationforthesubsequentlicenseplaterecognition.
II.IMAGEPREPROCESSING
Imagepreprocessingisanessentialprocessinlicenseplaterecognitionsystem,andthequalityofpreprocessingdirectlyaffectsthelocation.Theimagepreprocessinginthisarticleincludesimagegrey,edgedetection,medianfilterandbinaryzation.Thefollowinggivesthedetailstatementonthepreprocessingprocess.
A.Imagegreying
Allvehicleimagesacquiredthroughcameraandimagecardarecolorimage,andimageformatisnotsame.ThecommonlyusedimageformsareJPEGandBMP.Iftreatmentwithacquiredimagedirectly,notonlytheimageformatiscomplex,moreoverthecomputationdataquantityisextremelyhuge,suchlicenseplatelocationcannotsatisfytherequestforfastandreal-time.Therefore,thecolorimageneedtobeformattedprocessing,transformingtheJPEGimageortheBMPimagetoDIB(DeviceIndependentBitmap)whichfavorsthecomputertoprocess.Then,usingR,GandBtricolorweightedaveragemethodprocesstheDIBimage,processingfunctionisshownasequation
(1):
F(x,y)=0.299*R(x,y)+0.587*G(x,y)+0.114*B(x,y)
(1)
R(x,y),G(x,y)andB(x,y)areR,GandBtricolorcomponentoftheinputcolorimage[3].Thecolorimagetransformstogreyimagebyequation
(1)processing,theresultisshowninFig.1.
(a)Originalimage
(b)Greyimage
Figure1.Acontrastbetweenoriginalimageandgreyimage
B.Edgedetection
Edgeisthemostbasicfeatureoftheimage,sotheedgeindicatesstepchangeofthegreylevelonitssurroundingpixels.Chinesevehicleslicenseplateregionhasbigcolorcontrastbetweenlicenseplatebottomandlicenseplatecharacter.Thelicenseplateiscomposedof7characterswithrichedgeinformationinturnwhicharetheChinesecharacter,theletterandtheArabicnumeral,andthecharacterinlicenseplateregionandbackgroundhaveobviousedgeintheentirepicture,alsohavemanyedge.Thisisoneofthebasiccharacteristicsthatlicenseplateregiondistinguishesfromotherregioninthevehiclespicture,anditisalsothefundamentalbasisofthisalgorithm.CommonlyusededgedetectionoperatorshavePrewittoperator,Sobeloperator,Cannyoperator,LOGoperator,Robertsoperatorandotheroperator.PrewittoperatorandSobeloperatorarefirst-orderdifferentialoperator,theformeristheaveragefilter,thelatteristheweightedaveragefilter,whiletheimageedgedetectedbythetwomethodsmaybebetterthantwopixels.TheCannymethodusesfirstderivativeasthefoundationtojudgeedgepoints.Itisoneofthebesttraditionalfirst-orderdifferentialoperatorsinthedetectionofstepedge.Theshortcomingissmoothingoutsomedetails[4].LOGoperatorusesGaussianfunctiontosmoothimagefirst,thenusesLaplacetransformtoprocessimage,andthismethodprocessingimageedgeisinsufficientlyclearandthespeedneedstobeimproved.ThelocalizationusingRobertsoperatorisquiteprecise,butmoresensitivetonoise.TheexperimentindicatedthatusingthePrewittedgedetectionoperatorcanbetterstandouttheedgecharacteristicoflicenseplate,andspeedisfaster.Fig.2isseveralimagesafterprocessbydifferentedgedetectionoperator.
(a)Prewittoperator
(b)Sobeloperator
(c)Cannyoperator
(d)LOGoperator
(e)Robertoperator
Figure2.Imagecontrastafterprocessbyseveraledgedetectionoperator
C.Medianfiltering
Medianfilteringmethodisanon-linearsmoothingtechnique.Itsetsthegreylevelofeachpixeltothemiddlevalueofallpixels’greylevelinaneighborhoodwindow[5].Medianfilteringmethodisanon-lineartechniquethatbasedonasequencingstatistictheory.Itcaninhibitthenoiseeffectively.Thebasicprincipleofmedianfilteringistoreplacethevalueofpointindigitalimageornumericalsequenceswiththemiddlevalueofthispoint’soneneighborhood,soitcanletthearoundpixelvalueclosetothispoint’svalue.Thustheisolatednoiseiseliminated.Thismethodutilizesthetwo-dimensionalslidingtemplateofacertainstructure,arrangesthepixelintemplateaccordingtothesizeofpixelvalue,thenarise(ordrop)two-dimensionaldataarraywasproduced.Theoutputoftwo-dimensionalmedianfilteringresultprovidedbyequation
(2):
G(x,y)=med{F(x-k,y-l)}
(2)
F(x,y),G(x,y)isrespectivelyfororiginalimageandtheimageafterdealingwith.Wisatwo-dimensionaltemplate.TheresultisshowninFig.3.
Figure3.Medianfilterimage
D.Imagebinaryzation
Imagebinaryzationprocessingisthatsettinggrayvalueofpixelsonimageto0or255,thatis,theentireimagepresentstangibleblackandwhiteeffect[6].Weobtainthebinaryzationimagethroughselecting256brightnesslevelofgreyimagebysuitablethreshold.Binaryzationimagecanstillreflectthewholeandpartialcharacteristicofimage.Indigitalimageprocessing,binaryzationimageholdstheextremelyimportantstatus.First,binaryzationreducestheamountofimagedata.Secondly,tohighlighttheoutlineoftheinterestgoal,thisisfavortofurtherprocessing.Fig.4istheimageafterbinaryzationprocessing.
Figure4.Imageafterbinaryzationprocessing
III.LICENSEPLATELOCATION
Thetaskoflicenseplatelocationistoremovemostunwantedbackgroundinformationfromthewholeimageandfindthelicenseplateregionwithasmallamountofredundantbackground.Becausethelicenseplateregioncontraststothebackground,thehistogramoflicenseplateimageshowsabimodalshapeafterimagepreprocessing.Thewavetroughbetweentwowavepeakscorrespondingtothegraylevelisselectedasathreshold.SupposingtheimageisdividedbyF(x,y)andthegraylevelrangeis[Z1,Zk].Fig.5showstherearetwoobviouswavepeaksingraylevelZiandZj,andinZtthereisawavetrough.BychoosingZtreasonably,B1beltcancontaingreylevelcorrelationtothebackgroundasfaraspossible,whiletheB2bandincludesgreylevelcorrelationtothelicenseplateasfaraspossible[7].
Figure5.Doublepeakofhistogram
A.Locatingupperandlowerboundary
Onecharacteristicoflicenseplateimageisthecrowdedcharactersintheinternal,sothegreyjumpisextremelyfierce.Wefindthepossiblelocationoflicenseplateregionbyusingrowgreyjumpruleofgreyimage.Wepreservethispositionandcallitasthefakelicenseplateregion.Specificalgorithmincludingfollowingsteps:
Step1:
Calculatingthelevelhistogramofimage,andsmoothingthelevelhistogramwith[1,1,1,1,1]/5operator.
Step2:
Searchingthebottomedgedistanceoflicenseplatefromtheimagebase,if5linewhichispredefinedas5hassatisfiedtherequestcontinuouslywhichthevalueisbiggerthanorequalto10pixelsinhistogram,andthevalueofcurrentlinediffersabove4pixelswiththevalueoffronttheNthline,thenwebelievethatthebottomedgedistanceoflicenseplatehasfounded.Currentlineminus5,andlocatesthescanlinetothesummitofcurrentpeak.Ifthecurrentlinedoesnotsatisfythecondition,thencontinuestosearchupwardlyuntilthetopmarginofimage.
Step3:
Locatingthecurrentlinetothebottomofupwavecrest,ifthepeakbottomvalueisgreaterthanthemaximumvalue,thenlocatingtothesummitofcurrentpeak,andthesummitformaximumvalueline;
searchingupwardlyfromthecurrentline’snextline,ifthevalueofsearchlineisgreaterthantherecordedmaximumvalue,thensettingthecurrentlineasmaximumvalueandcarryingonsearchingupwardlyfromit.Otherwise,ifthecurrentvalueissmallerthantwo-thirdsofmaximumvalue,orthecurrentvalueis