Topic: Towards Deep Generative Models for improving generalisation capabilities on Face Presentation Attack Detection
by Lazaro Janier Gonzalez-Soler
online Big Blue Button Room: D19/2.03a, February 04, 2021 (Thursday), 12.00 noon
Keywords — Presentation Attack Detection, Face, Generative Models
The large development experienced by social networks has unveiled security concerns related to potential attacks on biometric systems. In particular, several examples have shown how a non-authorised subject can easily download a photo or video of a given person and use it to gain access to numerous applications. This is an attack presentation which could be created using an unknown species or captured under a challenging environment condition. In order to address those security threats, many techniques have been employed for facial Presentation Attack Detection (PAD). Most of them focused on traditional end-to-end deep learning architectures due to their success in several computer vision and pattern recognition applications. A rather limited number of works have addressed challenging unknown attacks through generative models. In order to exploit their generalisation capabilities, we benchmark both traditional and deep generative models for facial PAD. The experimental results on challenging scenarios show that the latter are more suitable for the detection of unknown attacks.