Continuous multi-biometric user authentication fusion of face recognition and keystoke dynamics

Abstract

Several application scenarios require the user to be authenticated not only at the time of logging in to a device, but continuously, such as a mobile device being used for an extended period of time, or examinees attempting for an online test. In this paper, two widely used unimodal biometric systems, which can both easily be captured on modern computing devices, keystroke dynamics and face recognition, are fused to create a stronger multi-biometric system for continuous authentication. The matching score for keystroke dynamics system is obtained using nearest neighbor classification (combined distance) and for the face recognition system, the EigenFace approach is used. The fusion of matching scores obtained by these unimodal biometric systems at the score level improves the accuracy. Scores obtained from each individual biometric system on the CMU keystroke dynamics database and the ORL face database is normalized using min-max normalization before fusion. The sum, product and weighted sum rules have been used for fusion and the experimental results confirm that a multi-factor authentication system gives better accuracy than a single-factor authentication system. The experiments also indicate that the weighted sum rule outperforms the sum and product rule method.