Continuous face authentication scheme for mobile devices with tracking and liveness detection

We present a novel scheme for continuous face authentication using mobile device cameras that addresses the issue of spoof attacks and attack windows in state-of-the-art approaches. Our scheme authenticates a user based on extracted facial features. However, unlike other schemes that periodically re-authenticate a user, our scheme tracks the authenticated face and only attempts re-authentication when the authenticated face is lost. This allows our scheme to eliminate attack windows that exist in schemes authenticating periodically and immediately recognise impostor usage. We also introduce a robust liveness detection component to our scheme that can detect printed faces and face videos. We describe how the addition of liveness detection enhances the robustness of our scheme against spoof attacks, improving on state-of-the-art approaches that lack this capability. Furthermore, we create the first dataset of facial videos collected from mobile devices during different real-world activities (walking, sitting and standing) such that our results reflect realistic scenarios. Our dataset therefore allows us to give new insight into the impact of user activity on facial recognition. Our dataset also includes spoofed facial videos for liveness testing. We use our dataset alongside two benchmark datasets for our experiments. We show and discuss how our scheme improves on existing continuous face authentication approaches and efficiently enhances device security.