基于BP网络的人眼组织切片图像角膜分割

476()Vol.47No.6200811JournalofXiamenUniversity(NaturalScience)Nov.2008BP,3,(,361005):2008204218:(60371012,60601025,30770561),2(WKJ2005222001)3:[e-mail]:,.BP(Back2propa2gation),.,1/2,,,6,BP.,,.,,,.:BP;;;:TP391.41:A:043820479(2008)0620912204.,..[1],[2].35m,HE,,64613441500,BMP[3].;;..Can2nyFastmarching().,BP.1,BP,,BP,,,.1.11/2,1/2.13441500,BP,BP[4-5]23.1.1Fig.1Humaneyetissueimage,,1/2.,1/2,,,,R150,G20,B20.,().R150,G20,B20.:(1)columI,row.,(i,j)R150,G20,B20;(2)(i,j)33R150,G20,B20.,x0=i,y0=j,(1);(3)y014colum-50,I(1irow,1jcolum/2),flag=1;y034colum+50,I(1irow,col2um/2jcolum),flag=2,x012row,I(1irow/2,1jcol2um),flag=3;x012row,I(row/2irow,1jcolum),flag=4.1/2,.1.2BP1.2.124RGB,RGB0255.,,R=RR+B+G,G=GR+B+G,B=BR+B+G,,RGB01.1.2.2.RGB(Redgreenblue),.HSI(Huesaturationintensity),H,;S,;I,.HSI..HSI.RGB,,.,..,BPHSI,.,C=0.11r+0.59g+0.3b,RGB.33BP.RGB.3BP.,RGB,RGBBP.BP6,RGB33.1.2.36BP,6.2:.1,0,1.,,.,,,.MarenAJ,J=sqrt(m+n)+a.,m,n,a110[6].,.,4.1.2.4BPBP.:,,,.BP,.BP[7].BP,6,4,1.10.,753196:BP2BPFig.2ResultofBPnetworksegmentation,75.,.2.1.3.2BP,.,...1/2,,(910)(90%)1,,,.:(1)flag=1,,1A;flag=2,,1A;flag=3,;flag=4,.(2)A910185,,A,(1).(3),3.(4),4(1/2).,.24,.646.,,.3Fig.3Resultofregiongrowingandholesfilling4Fig.4Finalsegmentationresult.,1/2.23,BP.10,,90%,.BP.3,BP.,,BP,,,.,:,.:[1],,,.[J].,2007,19(5):1049-1052.[2],,,.[J].,2004,23(3):193-198.419()2008[3]XieJiezhen,WangBoliang,JuYing,etal.Researchonvir2tualchinesehumaneyeanditsapplication[C]PPProceed2ingsofthe200527thAnnualConferenceinMedicineandBiologySociety.Shanghai:IEEE,2005:2906-2909.[4],,,.BP[J].,2001,17(3):146-148.[5],.:I.BP[J].,2004,9(4):65-68.[6],,,.BP[J].,2007,34(3):234-236.[7],,,.MATAB[M].:,2003:276-278.CorneaSegmentationinHumanEyeTissueSliceImagesBasedonBPNeuralNetworkHULi2ling,WANGBo2liang3,JUYing(SchoolofInformationScienceandTechnology,XiamenUniversity,Xiamen361005,China)Abstract:Corneaisoneofthemostimportantpartsofhumanbodyseyes.Accordingtotheparticularobtainmentofthehumaneyestissueimages,inthispaper,amethodtoextractcorneaslocalinformationbasedonback2propagation(BP)neuralnetworkisproposed.First,wecapturehalfoftheimagesincludingcorneainformation.Thenseveralimageswithdifferentcorneacolorsarecho2senastrainingsamples.Manualsegmentationimagesoftrainingsamplesareusedasteachersignals.Thenetworksinputvectorsare6imagefeatures.WecantraintheBPneutralnetworkwith3layersandcapturethesegmentedimages.Finally,wegettheexactlycorneainformationbyregiongrowingwiththesegmentedimagesandfillingholesofimages.Thismethodisturnedouttogetexpec2tedpurposeandoperatelittlesubjective.Itisareferencemethodtosegmenttheotherpartsofeyesofthesetissuesliceimages,andalsoisabasisfortheresearchofmodalmodelofcornea.Keywords:BPneuralnetwork;cornea;imagesegmentation;humaneyesslice5196:BP