Classification of Fingerprint Images

Automatic fingerprint identification is one of the most important biometric technology. In order to efficiently match fingerprints in a large database, an indexing scheme is necessary. Fingerprint classification, which refers to assigning a fingerprint image into a number of pre-specified classes, provides a feasible indexing mechanism. In practice, however, large intraclass and small interclass variations in global pattern configuration and poor quality of fingerprint images make the classification problem very difficult. A fingerprint classification algorithm requires a robust feature extractor which should be able to reliablely extract salient features from input images. We present a fingerprint classification algorithm with an improved feature extraction algorithm and a novel classification scheme. This algorithm has been tested on the NIST-4 fingerprint database. For the 4,000 images in this database, error rates of 12.5% for the five-class problem and 7.7% for the four-class problem have been achieved. With a 20% reject rate (which eliminates most of the poor quality images in the database), the error of the four-class problem drops to 2.4%.