简介:Thefingerjointlinesdefinedasfingercreasesanditsdistributioncanidentifyaperson.Inthispaper,weproposeanewfingercreasepatternrecognitionmethodbasedonLegendremomentsandprincipalcomponentanalysis(PCA).Afterobtainingtheregionofinterest(ROI)foreachfingerimageinthepreprocessingstage,LegendremomentsunderRadontransformareappliedtoconstructamomentfeaturematrixfromtheROI,whichgreatlydecreasesthedimensionalityofROIandcanrepresentprincipalcomponentsofthefingercreasesquitewell.Then,anapproachtofingercreasepatternrecognitionisdesignedbasedonKarhunen-Loeve(K-L)transform.ThemethodappliesPCAtoamomentfeaturematrixratherthantheoriginalimagematrixtoachievethefeaturevector.Theproposedmethodhasbeentestedonadatabaseof824imagesfrom103individualsusingthenearestneighborclassifier.Theaccuracyupto98.584%hasbeenobtainedwhenusing4samplesperclassfortraining.Theexperimentalresultsdemonstratethatourproposedapproachisfeasibleandeffectiveinbiometrics.