Automaticlatent fingerprint identification isausefultoolforcriminalinvestigation.However,the
accuracy ofidentification reportedinthestate-of-the-artliteratureislowduetothedistortioninlatent
fingerprint images.Inthispaper,wedescribeanewalgorithmbasedontheuseofclusteringwhichis
independent oftheminutiaedescriptors.Theproposedtechniqueimprovestherobustnessofidentification
inthepresenceoflargenon-lineardeformationwhichisassociatedwithlatent fingerprint images.
The newalgorithm finds multipleoverlappingclustersofmatchingminutiaepairswhicharemerged
togetherto find matchingminutiae.Severalexperimentsperformedusinglatent fingerprintdatabases
show thatourproposedalgorithmachieveshigheraccuracythanthosepresentedinstate-of-the-art
literature.Moreover,theresultsshowthattheproposedalgorithmissuccessfulindealingwiththelarge
distortion associatedwithlatent fingerprints formedundertheworstconditions.