Adaptive multimodal biometric fusion algorithm using particle swarm optimization and belief functions

In this paper, an adaptive multimodal biometric fusion algorithm is proposed. It is based on belief functions and Particle Swarm Optimization (PSO). The fusion is performed at the score level using belief functions such as Dempster Shafer, Yager, Proportional Conflict Redistribution and Dezert-Smarandache hybrid rules. A hybrid PSO is employed to select the best belief function and estimate its parameters. Several experiments have been conducted on BANCA dataset and a comparison between the well established methods has been performed. The preliminary results provide adequate motivation towards future research in the application of optimization techniques in the belief functions.