Enhancing security through a hybrid multibiometric system

Biometric user authentication techniques for security and access control have evoked an enormous interest by science, industry and society in the last two decades. Scientist and researchers have constantly pursued the technology for automated confirmation of the identity of subjects based on measurements of physiological or behavioral traits of humans. But even the best single biometric system suffers from spoof attacks, intra-class variability, noise, susceptibility etc. To address this issue, we develop a hybrid multibiometric system which integrates multi-algorithm and multi-modal approaches of multibiometric system and use bilevel fusion to combine biometric information. We use face, ear and signature biometric traits which are first classified by three classification techniques- multilayer perceptron, Fisher-image and Bayesian network. The outcomes of these classifiers for face are fused by rank fusion method. Outcomes for ear and signature are also fused similarly. The second level fusion occurs when we combine the results of these three rank fusion methods' outcomes for face, ear and signature with decision fusion method. We use Borda count and Borda fuse approaches for rank fusion and majority voting, weighted majority voting and behavioral knowledge space approaches for decision fusion. The final results indicate that this hybrid multi biometric system outperforms the single biometric systems build on the same data using the same classification algorithms. This system can be effectively used in law enforcement or homeland security department or for commercial purposes.