Optimal score level fusion combining multi-normalisation and separability measures

Abstract

This paper demonstrates the utility of multi-normalisation and separability measures for the optimal fusion of fingerprint and speaker biometrics. The decision scores of the individual matchers are transformed using various normalisation techniques and the global scores are obtained by combining the multi-normalised scores using the weighted fusion rules. The class as well as the score separability measures, under various noise conditions are estimated and combined algebraically, to determine the best integration weight, for the complementary modalities employed. The weight factor is optimised against the recognition accuracy. Experiments done with chimeric user database result in minimising the intersection between the genuine and the impostor score distributions, which in turn reduces the classification errors. Hence, by incorporating multi-normalisation and integration weight optimisation scheme on a unified framework, we can achieve better recognition performance and make the system robust to fluctuating inputs, even under extreme noise conditions.