Fingerprint Identification with Exclusive and Continuous Classification

Fingerprint identification from large database is a great challenge as both the accuracy and the speed of one-to-many matching may deteriorate significantly comparing to verification where only one-to-one matching is needed. This paper proposes a combined classification approach, which takes advantages of exclusive and continuous classification by performing the continuous classification after the exclusive classification. To circumvent the problems of small number of classes in the traditional exclusive classification, a clustering technique is proposed that exploits the similarities among the database templates and partitions them into much larger number of non-overlapping clusters than the conventional exclusive classification. Clustering technique serves as a coarse level classification which speeds up the followed continuous classification. Experimental results on NIST-4 fingerprint database demonstrate that the proposed approach outperforms various existing approaches in terms of not only the classification accuracy but also the classification efficiency