Rank Based Hybrid Multimodal Fusion Using PSO

Abstract

This paper investigates a hybrid fusion methodology by incorporating ranks and decisions of a user in a multimodal biometric system using PSO. Each of the biometric modalities is first used to rank the users. The individual ranks are then integrated to get a fused rank for each user. The matching scores associated to the fused ranked users are employed to take accept or reject decisions for each modality. The final decision is made by integrating the two decisions by the individual modalities. The decision thresholds for two modalities and a decision level fusion rule are selected by incorporating PSO. The role of PSO is to adaptively choose the fusion parameters in the varying security needs by minimizing the error rates in the system. The proposed methodology has a particular importance when the scores associated with two modalities are in different domain and their integration on score level need extra complexity of normalization. The experimental results presented in this paper have shown the relevance of the proposed scheme.