Inthispaper,wepresentanewdeformablemodelforshapesegmentation,whichmakestwomodificationstotheoriginallevelsetimplementationofdeformablemodels.Themodificationsaremotivatedbydifficultiesthatwehaveencounteredinapplyingdeformablemodelstosegmentationofmedicalimages.Thelevelsetalgorithmhassomeadvantagesovertheclassicalsnakedeformablemodels.However,itcoulddeveloplargegapsintheboundaryandholeswithintheobjects.Suchboundarygapsandholesofobjectscancauseinaccuratesegmentationthatrequiresmanualcorrection.Theproposedmethodinthispaperpossessesaninherentpropertytodetectgapsandholeswithintheobjectwithasingleinitialcontourandalsodoesnotrequirespecificinitialization.Thefirstmodificationistoreplacetheedgedetectorbysomeareaconstraint,andthesecondmodificationutilizesweightedlengthconstrainttoregularizethecurveunderevolution.Theproposedmethodhasbeenappliedtobothsyntheticandrealimageswithpromisingresults.
