Fusion in Fingerprint Authentication: Two Finger Scenarios

This paper presents our initial results of fingerprint fusion by analysing all pair combinations of ten fingers (i.e. 45 pairs). As a fingerprint verification algorithm we use a commercially available software package from Neurotechnology. As a test data set we use GUC100 multi scanner fingerprint database which contains fingerprint images of all ten fingers from 100 subjects collected over several months. We conduct score level fusion and apply two fusion methods, namely max and sum rules. Our analysis indicates that fusion using both rules provide improvement in all scenarios, and max rule showing slightly better performance compared to max rule's performance. The obtained performance improvements (in terms of EER) for sum and max rules are in the range of 60.2%-91.1% and 67.4%-93.5%, respectively.