二维Otsu自适应阈值选取算法的快速实现

33920079ACTAAUTOMATICASINICAVol.33,No.9SePtember,2007otsu,22otsu,),._Otsu,,,.,_otsu,.,,,otsuTN911.73AFstAlgorithmforTwo-dimensionalotsuAdaPtiveThresholdAlgorithmWANGHaihngl,2PANDe-LulADe-ShenZSahoo[,otsu.,,,,,.,.,[zlOtsu.,,,fOtsu.,,P42.8GCPU,256..,.,,,.,,.2OtsuL,L,,:j,(,j,,,:l)AbstractAsaelassiealimageseglllentionmethod,OtsuadaPtivethresholdalgorithmhasbeenaPPliedwidelyinimageProcessing.Butthelong-timeeomPutationlimitstheuseofthetwo-dimensionalOtsuthresholdajgorithmbasedontheOtsuthresholdalgorithm.Thpapergivesafastalgorithmfortwo-dimensionalotsuadaPtithresholdalgorithmthatovereomesthedisadvantageofeomPationaleomPlety.Thisfastalgo-rithmgetsridofredundantcomPutationandeldsalookuPtablebyiteration.TheexPerimentalresulthasdemonstredthattheeomputationaltimeofthefastmethod15notonlyfarlessthanthatoftheorigintwo-dimensionalone,butalsoyieldsthesaxnethresholdasthatoftheoriginalmethod.KeywordsThresholding,between-elassseatter,imageseg-memation,Otsu,j=,/N,j=1,2,,LN,N=:=--0lJ,t=--olE,j=1,,,.,,.,,.,,.],.,,.otsu[s],.1_Fig.1Four-quadrantpartitionofthetwdimeionalhisto,amplane20063292006-6-15ReeeivedMareh29,2006:inrevisedformJune15,20061.3100122.2100941.StateKeyLaboratoryofSatelliteoeeanEnvironmentDynamies,SeeondIntituteofoeeanograPhy,StateoeeanieAdministration,Hangzhou3100122.DePartmentofComPuter,NanjingUniver-sityofSeieneeandTeehnology,Nanjing210094DOI:10.1360/~007-0968C0(l)Cl(IH),(1),,(s,t),w0Pi,,W0(s,t)(2)9:Oteu969=,IJ=(s,t)J)WlJ!,QqJ4111.1f.(,),,=OJ=0,,)T(,J/W0,,Pi,,/w0)TOJ=0=OJ=0(4)X(,),,=oj=0,=(1,1,)T=(/Wlm,n)=J,,Ž=s+lj=t+1=s+IJJ,,/Wl)T(5)=OJ=0X0X(s,)Y0Y(s,t)(15)(16)LILILIL1s)0)111..O..rJr.(,J)T(,J,J,,)T(6)=oj=0=OJ=0Xl=X(L1,Ll)X(L1,X(s,L1)+X(,t),j,C0,Cl1IIn!,nIV,j,()Yl=Y(L1,Ll)y(L1,Y(s,L1)+Y(s,t)Wo(,s(s,t)=WO(s,)()T(T)+(s,)(,)T(:)=W0()T()+Wl(:)T(;)=W0(:)2+(,,)2+W1(1)2+(:,)2]Wl=(L1,Ll)(L1,t)p(s,L1)+p(s,t)r7______._____(X(,),Y(,n))(15)(p(,))~(20),.(s,t)P(,)=P(1,)+P(,l)P(1,1)+(21)s(s,)=max{S,(s,t)}3otsu(4)(5)P(o,0)=X(,)=X(1,n)+X(,n1)X(1,n1)+x(22)=(,J)T=(E,/w0,JPi,,/W0)T==oj=0=OJ0(,,,Ej,,)T/WoX(0,0)=oxPo.(8)Y(,)=Y(1,n)+Y(,n1)Y(1,n1)+x(23)=OJ=0=oj=0Ul=L11(i,)T,j=s+lj=+1LILl/Wl,L1J,j/WI)T==lj=+1L1(9)(E,j,EJ,,)T/Wl=+IJ=+1=s+IJ=t+1X0=E,J,Y0==oj=0EJ,j,=oj=0LIL1Y(o,o)=oxP0o4L,Otsu,,,8.8,,3+3(L3)(Lt),Xl=L1E=,+ljPi,j,Yl=E=s+IJEJ,j.=t+1LIL1(,)T=(xo,YO)T/(;,1,)T=(xl,h)T/AI[35+3(Ls)(L)]O(L)(24)s=ot=0,,,9,(21)(23),1ƒUl..1”.†97033(Ll)x(L1)xg,2()Table2performaneeeomparison(portableeomputer)LIL1A:gxLxL+918xLxL18xL2=o(2)(25)(s)(s=ot=0,.(,,.,2.(90,147)(66,140)(141,89)46.7546.7546813(90,147)(66,140)(141,89)0160.0150.015a)c)(a)(a)Bloc,L.f,,3r;4r.3(128),,(128),,;1024(256),0.594,3.145728MB(4),,,.3rble3()Timeeost(s)6412825651210240.0930.0011.6090.015263750031453.4840.01257017860.594))Camera(e)(e)ce4ble4()Spoeost(Byte6412825651210242Fig.2ExPerimelltaiimages122884915219660878643231457281P42.8GCPU512MVC6.0,1000,.1(7),.,,,,.2PM2.0GCPUIGvC6.02,.5Oteu,,.,,0()0(),,.1()lhble1Performanceeomparn(desktoPeomputer)(s)(s)(90,147)(66,140)(141,89)49.53249.18849.078(90,147)(66,140)(141,89)0.0470.0320031(a(c,1SonM,HlavacV,BleR.IZng~g:AnjsandMjneson.Beijing:Pts&TelecomPr,20()32Liujian-Zhuang,LIWnQing.Theautomatletliresholdingof.levelPictureviaZDotsumethod.AetaAutoznacaj,1993,19(1):101105(,.Otsu.,1993,19(1):101105)3OtsuN.AtliresholdseleCtionmethodfromgry.levelhls-a.EEansactionsonSysterns,Mn,and(ber-neC8,1979,9(1):9199264SahooPK.Asujveyofthresholdtechques.puter~sionGrapie,IinaPro~,1988,41(2):23326033920079ACTAAUTOMATICASINICAM1.33,No.9SePtember,20076WUS,AminA.Automatiethresholdingofgraylevelusingmultistageapproaeh.In:ProeeedingsofIEEEInternationalConfereneeonDoeumentAnaj15andReeognition.IEEE,2003,123812427TsaiDM.Afastthresho1dingselectionProcedureformul-timodalandunimodalhisto,ams.PaternReCognjLet-ers,1995,16(6):6536668YinPY,ChenLH.Anewmethodformultilevelthreshold-ingusingsymmetryanddualityofthehistogram.In:Preeedinof1994InternationalSymposiumonSPeeeh,ImageProeessingandNeuralNetworkS.IEEE,1994.1:45489DavidJE,AlbertTA,Cheno,vethDL.TvCdimensionalentroPieseglllentation.PatemRecjjonLeers,1999,20(3):32933610AlbuquerqueMP,EsquefIA,GesualdiMAR.ImagethresholdingusingTsailiseDtroPy.PaernRecjjonLe-ers,2004,25(9):1059~106511SahooPK,AraraG.Athresholdingmethodbedontwo-dimensionalRenyi,5entroPy.PaernRecognjtion,2004,37(6):1149~1161.,,..E-maiwhyyang@126com(WANGHaiyngPh.D.eandidate.Hisresearchinter-esteoversimeProcessing,eomPutervision,andPatternreeognltion.Correspondingauthorofthisper.)111,.,,,,.,,TP391.11ADefeetDeteetionAlgorithmforPaPerCurreneySortingSystemJINYelSONGLingiLIUSonBol,ANGXiang-LOngl.,,.E-mail:[e-mail](PANDe-LuProfesor.Hisresearchinteresteovrsre-motesensingofoeeaneolor,satelliteremotesenseimageanalyzing,andpatternreeognition)AbstraetThispaPeradVaneesahomogeneitybedalgrithmforthedetectionofserateheanderacksaPPearingonPaPereurrenCy.Thehomog:eneityfeatureofthesenedPSPereurrencyimage15firstconstruetedtoloeatethePossiblyPol-lutedPixels.Theimagerestrationaigorlthm15subsuentlyusedtooverlthesensedandreferencepapererencyimagesAtlast,eaChpossiblypollutedpixelonth