Multi-Modal Biometric Verification Based on FAR-score Normalization

Fusion different biometrics is an effective way to design a biometric system with robust performance. To do this, normalization functions are employed. However, these functions can not follow the distributions of scores from distinct classifiers. Consequently different normalization errors are introduced. In this paper, the scores from different classifiers are converted into the corresponding false accept rate (FAR), which introduces smaller normalization error than traditional methods, and makes the fusion more operable. To further enhance the fusion result, a dynamic selection of fusion rule is implemented based on the discrepancy between scores of different classifiers. Experiments conducted on a multi-modal biometric system composed of face and fingerprint verification system show our methods are superior to conventional approaches.