Fingerprint Preselection Using Eigenfeatures for a Large-Size Database

We describe fingerprint preselection using eigenfeatures for automated fingerprint identification systems in large-size, 10-print card databases. As preprocessing for minutiae matching, the preselection reduces minutiae matching candidates. Eigenfeatures and confidence factors are extracted for the preselection from the fingerprints in the 10-print cards. The eigenfeatures are calculated by the KL (Karhunen-Loeve) expansion of the ridge directional pattern of the fingerprint, and the confidence factors are extracted as confidence of the eigenfeatures. Using the confidence factors, we can adaptively calculate a matching distance according to variances of the features under the confidence. This adaptation significantly improves preselection accuracy and achieves 1/1,500 to 1/10,000 reduction of minutiae matching candidates. The average number of instructions needed to compare a pair of cards is approximately 90 and is 105 times faster than minutiae matching computation. This preselection scheme en ables the realization of fingerprint identification systems of high computational performance.