Score-level fusion for cancelable multi-biometric verification

Integration of scores from multiple biometric modalities has become promising to alleviate the limitations of unibiometric systems such as sensitivity to outliers, erroneous authentication caused by inter-class and intra-class variability and low verification performance due to poor quality. In this work, we propose a two-level score level fusion approach for integrating the scores obtained from cancelable templates of different biometric modalities. As a result, we achieve a significant improvement in overall recognition performance providing secure authentication for the different application. At the first level, scores from multiple matchers are combined using a novel mean-closure weighting (MCW) technique to achieve the desired score for a particular biometric modality. The proposed solution is based on the region of uncertainty between the genuine and imposter distribution. Further, the derived scores from different modalities are integrated using a novel rectangular area weighting (RAW) technique at the second level to obtain the overall fused score. Overall, the proposed two-level cancelable score fusion method improves the performance over unimodal cancelable systems and are more robust to the variability of scores and outliers. The evaluation has been performed on two virtual databases and is compared with the existing weighting, density and classification based score fusion techniques. Experimental results show that the proposed two-level cancelable score fusion improves the overall performance over unibiometric system satisfying the requirement of secure authentication. (C) 2018 Elsevier B.V. All rights reserved.