LAUSR

The curse of dimensionality is a well-known problem in biometric applications (e.g., biometric passports). The downside of this problem is that both the accuracy and speed of the biometric authentication process are reduced. This paper sets forth a feature selection (FS) method based on speed-constrained multi-objective particle swarm optimization (SMPSO). The proposed approach aims to reduce the size of the biometric features through the minimization of the intra-class variations and the maximization of the inter-class variations. Experiments have been conducted using several datasets from University of California–Irvine (UCI) to confirm the efficiency of SMPSO-based FS against state-of-the-art multi-objective FS approaches, such as the multi-objective evolutionary algorithm based on decomposition (MOEA/D) and the second non-dominated sorting genetic algorithm (NSGA-II). When compared to NSGA-II and MOEA/D, SMPSO gained 6.01% and 6.11%, respectively, in average classification accuracy. Moreover, SMPSO achieved the best accuracy compared to MOGA, a modified version of NSGA-II. The experimental results obtained by using a YALE Face database validated the effectiveness of the proposed approach in reducing the size of the biometric features while allowing a good recognition accuracy. The classification performance was improved by 8.2% compared with the performance of the stateof-the-art approaches.