A new adaptive combination approach to score level fusion for face and iris biometrics combining wavelets and statistical moments

Abstract

In this paper, we present a new combination technique to fuse scores deriving from face and iris biometric matchers. Based on a precise statistical analysis of bootstrapped match scores deriving from similarity matrices, we show the utility of wavelet denoising on normalized scores. Then, we use an adaptive fusion rule based on the maximization of a cost function combining user-specific weights, a separation distance and statistical moments. Experiments are conducted on FERET and CASIA databases and results show that our proposed method outperforms by 70% some of the best current combination approaches in terms of Equal Error Rates (EER), and reaches a Genuine Accept Rate (GAR) equals to 100% at a False Accept Rate (FAR) of 7×10 -∈4%. © 2008 Springer Berlin Heidelberg.