摘要
Itiswidelyacceptedthatthedesignofmorphologicalfilters,whichareoptimalinsomesense,isadifficulttask.Inthispaperanovelmethodforoptimallearningofmorphologicalfilteringparameters(Genetictrainingalgorithmformorphologicalfilters,GTAMF)ispresented.GTAMFadoptsnewcrossoverandmutationoperatorscalledthecurvedcylindercrossoverandmaster-slavemutationtoachieveoptimalfilteringparametersinaglobalsearching.Experimentalresultsshowthatthismethodispractical,easytoextend,andmarkedlyimprovestheperformancesofmorphologicalfilters.Theoperationofamorphologicalfiltercanbedividedintotwobasicproblemsincludingmorphologicaloperationandstructuringelement(SE)selection.Therulesformorphologicaloperationsarepredefinedsothatthefilter'spropertiesdependmerelyontheselectionofSE.Bymeansofadaptiveoptimizationtraining,structuringelementspossesstheshapeandstructuralcharacteristicsofimagetargets,andgivespecificinformationtoSE.Morphologicalfiltersformedinthiswaybecomecertainlyintelligentandcanprovidegoodfilteringresultsandrobustadaptabilitytoimagetargetswithclutterbackground.
出版日期
2003年01月11日(中国期刊网平台首次上网日期,不代表论文的发表时间)