Fast computation of the performance evaluation of biometric systems: application to multibiometric

TL;DR

This paper introduces a fast approximation method for computing the Equal Error Rate (EER) in biometric systems, significantly reducing computation time and enhancing the efficiency of system evaluation and parameter optimization.

Contribution

It presents a novel, efficient approximation technique for EER calculation, facilitating quicker performance evaluation and parameter tuning in multibiometric systems.

Findings

The proposed method reduces EER computation time substantially.

It improves the efficiency of non-parametric confidence interval estimation.

It accelerates genetic algorithm-based parameter learning for biometric fusion.

Abstract

The performance evaluation of biometric systems is a crucial step when designing and evaluating such systems. The evaluation process uses the Equal Error Rate (EER) metric proposed by the International Organization for Standardization (ISO/IEC). The EER metric is a powerful metric which allows easily comparing and evaluating biometric systems. However, the computation time of the EER is, most of the time, very intensive. In this paper, we propose a fast method which computes an approximated value of the EER. We illustrate the benefit of the proposed method on two applications: the computing of non parametric confidence intervals and the use of genetic algorithms to compute the parameters of fusion functions. Experimental results show the superiority of the proposed EER approximation method in term of computing time, and the interest of its use to reduce the learning of parameters with genetic algorithms. The proposed method opens new perspectives for the development of secure multibiometrics systems by speeding up their computation time.