Fingerprint classification isstillachallengingproblemduetolargeintra-classvariability,smallinter-class
variabilityandthepresenceofnoise.Todealwiththesedifficulties, weproposearegularizedorientation
diffusion modelfor fingerprint orientationextractionandahierarchicalclassifier for fingerprint
classification inthispaper.Theproposedclassification algorithmiscomposedof five cascadingstages.
The first stagerapidlydistinguishesamajorityofArchbyusingcomplex filter responses.Thesecond
stage distinguishesamajorityofWhorlbyusingcorepointsandridgeline flow classifier.Inthethird
stage, K-NN classifier finds thetoptwocategoriesbyusingorientation field andcomplex filter responses.
In thefourthstage,ridgeline flow classifier isusedtodistinguishLoopfromotherclassesexceptWhorl.
SVM isadoptedtomakethe final classification inthelaststage.Theregularizedorientationdiffusion
model hasbeenevaluatedonaweb-basedautomatedevaluationsystemFVC-onGoing,andapromising
result isobtained.Theclassification methodhasbeenevaluatedontheNISTSD4.Itachieveda
classification accuracyof95.9%for five-class classification and97.2%forfour-classclassification without
rejection.