A Modified Equal Error Rate Based User-Specific Normalization for Multimodal Biometrics

Previous studies have shown that the performance of a biometric authentication system can be further improved by normalizing the matching score for each claimed identity. These techniques are known as user-specific score normalizations. Following this vision, the proposed research focuses on developing a new user-specific score normalization procedure, which is based on a recently proposed EER-Norm. While in its original form, some parameters specific to a user cannot be estimated due to the limited availability of training data, especially of the genuine/client matching scores, we aims to stabilise the estimates of these parameters by using both the user-independent and user-dependent information. The proposed approach tested on the XM2VTS and BioSecure DB2 databases is shown to outperform the existing known score normalization ones, such as Z-, EER-, and F-Norms in the majority of experiments.