Agent-based image iris segmentation and multiple view boundary refining

《Agent-based image iris segmentation and multiple view boundary refining》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Agent-BasedImageIrisSegmentationandMultipleViewsBoundaryRefiningRuggeroDonidaLabati,VincenzoPiuriFellow,IEEE,FabioScottiMember,IEEEAbstract—Thepaperpresentstwodifferentmethodstodealwiththeproblemofirissegmentation:anagent-basedmethodcapabletolocalizethecenterofthepupilandamet

2、hodtoprocesstheirisboundariesbyamultipleviewsapproach.Inthefirstmethod,anagentcorrespondstothecoordinatesofaspecificpointofanalysisintheinputimage.Apopulationofagentsisdeployedintheinputimage,then,eachagentcollectslocalinformationconcerningtheintensitypatternsvisibleinitsregio

3、nofinterest.Byiterations,anagentchangesitspositionaccordinglytothelocalproperties,movingtowardstheestimationofthepupilcenter.Ifnoavailableinformationispresentinitsregionofinterest,theagentwillmoveitselfFig.1.MultiAgentPupilLocalization.Startingfromaninitialspatialconfiguratio

4、n(left),agentssearchforthepupilcenterbyaniterativealongarandomwalk.Afterfewiterations,thepopulationapproach.Whenthestoppingcriterionisreached,theareawiththehighesttendstospreadandthenconcentrateintheinnerportionofconcentrationofagentsintheimage(right)estimatesthepupilcenter.

5、thepupil.Oncethecenterofthepupilhasbeenlocated,theinnerandouteririsboundariesarerefinedbyanapproachbasedonmultipleviewsanalysis.Thismethodstartswithasetofpointsthatcanbeconsideredasanapproximationofthepupilcenter.Foreachpoint,adetailedestimationoftheirisboundariesiscomputed,a

6、ndthefinaldescriptionoftheirisboundariesisobtainedbymergingalltheobtaineddescriptions.ThetwomethodsweretestedusingCASIAv.3andUBIRISv.2images.Experimentsshowthattheproposedapproachesarefeasible,alsoineyeimagestakeninnoisyornon-idealFig.2.MultipleViewsBoundaryRefining.Startingfr

7、omasetofobservationpointsinsidethepupil(redcirclesintheleftsubplot),theinnerconditions,achievingatotalerrorsegmentationaccuracyuptoandouteririsboundaryarereconstructedbyprocessingtheradialgradient97%.transitionsforeachobservationpoint(coloredcurvesinthecentersubplot).Thediff

8、erentboundariesarefusedinordertoestimatetheirisedges(right).I.INTRODUCTIONT