Meanshift算法在带钢缺陷图像分割中的应用

:2007206210:(19782),,1Meanshift1,2,1,2(1.,150001;2.,201900):.,Meanshift,.,Meanshift,,.:;;;Meanshift:TP391.4:A:100122400(2007)0621015204ApplicationoftheMeanshiftalgorithminsteelstripimagesegmentationYANGShui2shan1,HEYong2hui2,ZHAOWan2sheng1,PENGTie2gen2(1.SchoolofMechanicalandElectricalEngineering,HarbinInstituteofTechnology,Harbin150001,China;2.ResarchInstituteofBaoSteel,Shanghai201900,China)Abstract:Defectsegmentationresultsaffecttheprecisionofclassificationintheautomaticstripsurfacedefectdetectionsystem.Inordertoobtainabetterresult,themeanshiftalgorithmispresentedtosegmentthedefectsinstripimages.Themeanshiftalgorithmisusedtosmooththeregionofinterestintheimageanddetectthedefectsedge,anditseffectivenessiscomparedwiththatofthemedianfilter.Testresultsshowthatitcansmoothstripimageseffectivelyandcandetecttheedgesofdefectobjectsaccurately.Themeanshiftalgorithmcanworkeffectivelyinstripimagesegmentation.KeyWords:imagesegmentation;steelstripdefects;imagerecognition;Meanshift,,.,(ROI),ROI.,.,,.,.Meanshift,,[1].,,.Meanshift,.,Meanshift,,,.,Meanshift.,,.1Meanshift1975,FukunagaHostetlerMeanshift[2],200712346()JOURNALOFXIDIANUNIVERSITYDec.2007Vol.34No.6,.,Comaniciu[35].Meanshift,,.,,..111dnxi,i=1,2,,n,xiRd.K(x),H,x^f(X)=1nni=1KH(x-xi),(1)KH(x)=H1/2K(H-1/2x).dK(x),RdK(x)dx=1,RdxK(x)dx=0,limxxdK(x)=0,RdxxTK(x)dx=cKI,(2)cK,I.K1(x),K(x)=ak,dK1(x),(3)RdK1(x),.ak,da-1k,d=RdK1(x)dX.(4)1,.(2).,[4].,ComaniciuK(x)=ck,dk(x2),(5)k(x),ck,d0K(x)1.,H.,HH=diag[h21,h22,,h2d]H=h2I.,h^f(x)=1nhddi=1Kx-xih.(6)(MSE)(MISE).,,(AMISE).,n,h0,nh.(5),(6)^fk,d(X)=ck,dnhddi=1kx-xih2.(7),.,¨f(X)0.,.,.112(7)d,,.,¨fh,K(x)=¨^fh,K(x)=2ck,dnhd+2di=1(x-xi)kx-xih2.(8)6101()34,x[0,),g(x)=-k(x),(9)G(X)=cg,dg(x2),(10)cg,d.(6),¨fh,K(x)=2ck,dnhd+2di=1(xi-x)gx-xih2=2ck,dnhd+2di=1gx-xih2di=1xigx-xih2di=1gx-xih2-x.(11)di=1gx-xih20.,,.(11)Meanshift,MS(x),MS(x)=di=1xigx-xih2di=1gx-xih2-X.(12)Meanshift,,.,.,MS(X)0,,.1Meanshift2,.,.Meanshift,,.,1.ROI,Meanshift.,Meanshift,ROI.2,,3ROI.1ROI.,Meanshift,h4.(12),xt2,Meanshiftxt+1.xt+1=MNi=1xigxt-xih2MNi=1gxt-xih2,(13)M,N.g(y)=1,y10,,g(y)ROIxth.MS(xn)=xn+1-xn,(14)(14),MS(xn)0,xn+1.,,.9(33)[6].3,Meanshift.,,Meanshift.Meanshift,.71016:Meanshift3Meanshift4MeanShift,Kirsch,4.,Meanshift,.,Meanshift.3Meanshift.:,MeanShift,Kirsch,,..:[1]PengTG,WuTH.MeanShiftAlgorithmEquippedwiththeIntersectionofConfidenceIntervalsRuleforImageSegmentation[J].PatternRecognitionLetters,2007,28(2):2682277.[2]FukunagaK,HostetlerLD.TheEstimationoftheGradientofaDensityFunction,withApplicationinPatternRecognition[J].IEEETransonInformationTheory,1975,21(1):32240.[3]ComaniciuD,RameshV,MeerP.TheVariableBandwidthMeanShiftandData2DrivenScaleSelection[C]//ProcofICCV2001.Vancouver:IEEE,2001:4382445.[4]ComaniciuD,MeerP.MeanShift:aRobustApproachTowardFeatureSpaceAnalysis[J].IEEETransonPatAnalandMachIntel,2002,24(5):6032619.[5]ComaniciuD,RameshV,MeerP.Kernel2BasedObjectTracking[J].IEEETransonPatAnalandMachIntel,2003,25(5):5642575.[6],,,.VisualC++[M].:,2001:2812283.(:)()8101()34