Fast learning for multibiometrics systems using genetic algorithms

The performance (in term of error rate) of biometric systems can be improved by combining them. Multiple fusion techniques can be applied from classical logical operations to more complex ones based on score fusion. In this paper, we use a genetic algorithm to learn the parameters of different multibiometrics fusion functions. We are interested in biometric systems usable on any computer (they do not require specific material). In order to improve the speed of the learning, we defined a fitness function based on a fast Error Equal Rate computing method. Experimental results show that the developed method provides very low error rates while having reasonable computation times. The proposed method opens new perspectives for the development of secure multibiometrics systems with speeding up their computation time.