Towards optimal countermeasures against wolves and lambs in biometrics
Takao Murakami, Kenta Takahashi, Kanta Matsuura · 2012
Claimants and enrollees who have high similarity scores against many others are referred to as wolves and lambs, respectively. These animals can cause false accepts against many others and compromise the security of the system. The aim of this paper is to develop a fusion algorithm which has security against both of them while minimizing the number of query samples the genuine claimant has to input. To achieve this aim, we first propose a minimum log-likelihood ratio based sequential fusion scheme (MLR scheme). We prove that the MLR scheme keeps WAP (Wolf Attack Probability) and LAP (Lamb Accept Probability), the maximum of the claimant-specific FAR and the enrollee-specific FAR, less than the desired value if the score distributions are perfectly estimated, except in the case of adaptive spoofing wolves. We also prove that the MLR scheme is nearly optimal with respect to the average number of inputs (ANI) of the genuine claimant under some conditions. We secondly propose an input order decision scheme based on the KL (Kullback-Leibler) divergence which maximizes the expectation of the genuine log-likelihood ratio, to further reduce ANI of the MLR scheme in the case where the KL divergence differs from one modality to another. The results of the experimental evaluation using the CASIA-FingerprintV5 showed the effectiveness of our schemes.