Remote Biometrics for Robust Persistent Authentication

This paper examines the problem of providing a robust non-invasive authentication service for mobile users in a smart environment. We base our work on the persistent authentication model (PAISE), which relies on available sensors to track principals from the location where they authenticate, e.g., through a smart card based access control system, to the location where the authentication is required by a location-based service. The PAISE model is extended with remote biometrics to prevent the decay of authentication confidence when authenticated users encounter and interact with other users in the environment. The result is a calm approach to authentication, where mobile users are transparently authenticated towards the system, which allows the provision of location-based services. The output of the remote biometrics are fused using error-rate-based fusion to solve a common problem that occurs in score level fusion, i.e., the scores of each biometric system are usually incompatible, as they have different score ranges as well as different probability distributions.

We have integrated remote biometrics with the PAISE prototype and the experimental results on a publicly available dataset, show that fusion of two remote biometric modalities, facial recognition and appearance analysis, gives a significant improvement over each of the individual experts. Furthermore, the experimental results show that using remote biometrics increases the performance of tracking in persistent authentication, by identifying principals who are difficult to track due to occlusions in crowded scenes.