A Novel Method for Fusion Operators Evaluating at Score-Level Fusion in Biometric Authentication

Biometrics is emerging as the most foolproof method of automated personal identification. And fusing the scores of several biometric systems is a very promising approach to improve the overall system's accuracy. Fusion operators, which contain sum rule, product rule, max rule and min rule, are considered to be one of the most useful schemes at score-level fusion, while the optimal fusion operator is chosen experimentally in real-world classification tasks. In this paper, a novel method is presented for optimal fusion operator selection. We estimate the PDF (probability density function) of each representation. Assuming that the representations used are conditionally statistically independent, then the PDFs of the fusion operators can be calculated. As a result, the distance between the class of genuine and impostor based on PDF can be used to evaluate the capabilities of fusion operators. It provides a theoretical support to evaluate the performances of fusion operators, and enables adaptive selection without experimentation. Its effectiveness when applied to bimodal biometric authentication is confirmed by the results of 21 experiments.