资源描述:
《基于学习的局部几何相似性的医学图像放大》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、ISSN1000-9825,CODENRUXUEWE-mail:jos@iscas.ac.cnJournalofSoftware,Vol.20,No.5,May2009,pp.1146−1155http://www.jos.org.cndoi:10.3724/SP.J.1001.2009.03546Tel/Fax:+86-10-62562563©byInstituteofSoftware,theChineseAcademyofSciences.Allrightsreserved.∗基于学习的局部几何相似性的医学图像放大1,21,2+3潘琪,罗笑南,朱继
2、武1(中山大学信息科学与技术学院,广东广州510275)2(教育部数字家庭重点实验室(中山大学),广东广州510275)3(广东药学院医药商学院,广东广州510006)Learning-BasedMedicalImageMagnificationAlgorithmbyLocalGeometricSimilarity1,21,2+3PANQi,LUOXiao-Nan,ZHUJi-Wu1(SchoolofInformationScienceandTechnology(SIST),SUNYAT-SENUniversity,Guangzhou510275,Ch
3、ina)2(KeyLaboratoryofDigitalLife(SUNYAT-SENUniversity),MinistryofEducation,Guangzhou510275,China)3(DepartmentofMarketing,GuangdongPharmaceuticalUniversity,Guangzhou510006,China)+Correspondingauthor:E-mail:lnslxn@mail.sysu.edu.cn,http://www.sysu.edu.cnPanQ,LuoXN,ZhuJW.Learning-Ba
4、sedmedicalimagemagnificationalgorithmbylocalgeometricsimilarity.JournalofSoftware,2009,20(5):1146−1155.http://www.jos.org.cn/1000-9825/3546.htmAbstract:Imagemagnificationisanimportanttechnologyinmedicalimageprocessing.Highdetailareasinmedicalimagesmostoftenhaveadefinitegeometric
5、structureorpattern,suchasinthecaseofedges.Thispaperproposesalearning-basedmethod.GeometricfeaturesextractedfromtheavailableneighboringpixelsintheLow-resolution(LR)imageformthetrainingset.AssumingthetrainingsetislocallycorrespondingtogeometricfeaturesfromtheHigh-resolution(HR)ima
6、gepatchtobereconstructed.Localgeometricsimilarityisdescribedasthecorrespondence.Thetaskofimagemagnificationisformulatedasanoptimizationproblem,wheretheoptimizationcoefficientscanadaptivelytuneitsvaluetoeffectivelypropagatingthefeaturesfromthetrainingsettothetargetHRimagepatch.Th
7、eadvantagesaretheabilitytoproduceamagnifiedimagebyanyfactor,andnotrequireanyoutliersupporters.AWeightedLeastSquare(WLS)methodisprovidedtoofferaconvenientwayoffindingtheregularizedoptimalsolution,wheretheweightfunctionisdeterminedbythenon-localmeans.Simulationandcomparisonresults
8、showthattheproposedmethodisindependent,adaptive