Fingerprint classification

A fingerprint classification algorithm is presented in this paper. Fingerprints are classified into five categories: arch, tented arch, left loop, right loop and whorl. The algorithm extracts singular points (cores and deltas) in a fingerprint image and performs classification based on the number and locations of the detected singular points. The classifier is invariant to rotation, translation and small amounts of scale changes. The classifier is rule-based, where the rules are generated independent of a given data set. The classifier was tested on 4000 images in the NIST-4 database and on 5400 images in the NIST-9 database. For he NIST-4 database, classification accuracies of 85.4% for the five-class problem and 91.1% for the four-class problem (with arch and tented arch placed in the same category) were achieved. Using a reject option, the four-class classification error can be reduced to less than 6% with 10% fingerprint images rejected. Similar classification performance was obtained on the NIST-9 database.