基于HSI空间和K-means方法的彩色图像分割算法

　27　720107MICROELECTRONICS&COMPUTERVol.27　No.7July2010:2009-06-12;:2009-08-12HSIK-means李丹丹,史秀璋(,100083)　:提出了一种新的彩色图像聚类分割算法,选用HSI空间作为彩色分割空间,在研究H分量的聚类算法中,该分量的圆循环特性被充分的考虑,同时也定义了H分量空间中两点距离的定义和中心的概念;选用最重要的H分量和I分量作为分割聚类特征,运用模糊隶属度刻画了该聚类特征,最后运用K-means算法对彩色图像进行聚类分割.实验结果表明,此算法能够准确地从彩色图像中提取目标区域,且在H分量和I分量上联合分割的结果好于在单个分量上分割的结果.:HSIK-means;H分量;聚类算法;模糊隶属度;彩色图像:TP31　　　　　:A　　　　　:1000-7180(2010)07-0121-04AKindofColorImageSegmentationAlgorithmBasedonHSISpaceandK-meansMethodLIDan-dan,SHIXiu-zhang(BeijingCityUniversity,Beijing100083,China)Abstract:Akindofclusteringalgorithmforcolorimagesegmentationisproposed,choosingHSIrepresentationasthecol-orsegmentationspace,inthestudyofHcomponent,thecirculationcharacteristicswerefullconsideration,alsothedis-tanceoftwopointsandcentersofHcomponentisdefined.ChoosingthemostimportantcomponentofHandIcompo-nentassegmentationclusteringfeatures,usingfuzzymembershipfunctiondepictstheclusteringfeatures,atlastK-meansclusteringalgorithmisusedtosegmentcolorimage.Anexperimentsisgivenonboneregionalsegmentation,andachievegoodsegmentationresults,experimentalresultsshowsthattheresultsoftheHandIcomponentsissuperiortotheresultsonasinglecomponentofHISspace.Keywords:HSI;K-means;Hcomponent;clusteringalgorithm;fuzzymembership;colorimages1　,,:[1]..,RGB.RGB,RGB.,RGB、CMYHSI.HSI,,HSIH()I()[2].(feature-space)(spa-tial-domain).,(RGBHIS),;,.,,,K-means、ISODATA、C-means、k-mediansClara[3],,K-means,,,,[4],[5],.HSI,K-meansHI,H,H,HI,K-means,.2　HSIK-means2.1　K-meansKK-.x=(x1,x2,…,xn)n,g(x)x,K-means[6]:(1)K,:μ1(1),μ2(1),…,μk(1).(2)i,K,x∈Q(i)l‖g(x)-μ(i)l‖‖g(x)-μ(i)j‖(1)Ql(i)il.(1).(3)i,μl(i+1):μi+1l=1Nl∑x∈Q(i)lg(x)(2),NlQl(i).(4)j=1,2,…,K,μ1(i+1)=μ1(i),;(2).2.2　HSIHSI,H,S,I,RGBHSIYC1C2=1/31/31/31-1/2-1/20-3/23/2RGB(3)HSI:I=Y(4)S=C21+C22(5)H=Arccos(C2/S)C1≥02π-Arccos(C2/S)C10(6)HSI1,,,H,S,IHS,1,HSI.1　HSI1,H,[0,2π).H,,,Hπ/3,5π/3,H.2.3　HSIH1HH,,H,H.K-means,K-means.H,HH1,H2:distance(H1,H2)=H2-H1|H2-H1|≤πH2-H1-2πH2≥H1,|H2-H1|≥π2π-H1-H2H1≥H2,|H2-H1|≥π(7)[H1,H2]HH1,H2:|H1,H2|={H≤|H1≤H≤H2}if|H1-H2|≤π(8)[H1,H2]={H|max(H1,H2)≤H≤2π}∪{H|2π≤H≤min(H1,H2)}if|H1-H2|≥π(2)HH1,H21222010.,,Hm:Hm=(H1+H2)/2|H1-H2|≤π(H1+H2)/2-π(H1-H2)≥π,(H1-H2)/2≥π(H1+H2)/2+π(H1-H2)≥π,(H1-H2)/2≤π(9),,X1,X2,…,XnHn,x1,x2,…,xn[H1,H2],Hm,X1,X2,…,XnXC=HM+1n∑ni=1distance(HM,Xi)(10)Euclidean(),Euclidean,,Hm,H,H,Hm.3　HSIHSI,S,HSIHI,:(1)S(2)S,H,H.(3)S,,I.,HI,,.,I,H,.,:,HSI,HI,.I,K-meansI,K-meansI,ImI.,I,IImI,1,ImI;0.5.I:μIO(x,y)=1-12|I(x,y)-mIC|μIB(x,y)=1-μIO(x,y)(11)μIO(x,y)(x,y)I,μIB(x,y)(x,y)I.,(x,y)IμIOμIB1.2.3,HH(K-means),HH,,I,:μHO(x,y)=1-12|H(x,y)-mOC|μHB(x,y)=1-μIO(x,y)(12)ΜHO(x,y)(x,y)H,μHB(x,y)(x,y)H.,(x,y)HμHOμHB1.,,IμIO(x,y)μIB(x,y),HμHO(x,y)μHB(x,y).,1,μIO(x,y)μHO(x,y)(x,y),,:F=(μIO(x,y),μHO(x,y))(13)K-meansF,.4　,2(a),,.2,2(a),2(b)I,2(c)H.123　7,:HSIK-means2(b),I,,I;2(c),.2(d)I,2(e)IH.2　,HSI,,,,.5　K-means,HSI,HI,H,HK-means,,K-meansIH.,,,.:[1]LuccheseL,MitrayS.Colorimagesegmentation:astate-of-theartsurvey[C]//ProceedingsoftheIndianNa-tionalScienceAcademy.California,2001:207-221.[2]HuangZK,LiuDH.SegmentationofcolorimageusingEMalgorithminHSVcolorspace[C]//ProceedingsofIEEEInternationalConferenceonInformationAcquisition.Seogwipo-si,2007:316-319.[3],..[M].2.:,1999:235-247.[4]ValavanisKP,ZhangJ,PaschosG.Atotalcolordiffer-encemeasureforsegmentationinColorimages[J].JournalofIntelligentandRoboticSystems,1996(16):269-313.[5].:,[M].2.:,2005:100-101.[6].:MATLAB,[M].:,2007:153-154.:　,(1972-),,..　..1242010