Topic: Towards Generalisable Facial Presentation Attack Detection Based on the Analysis of Facial Regions
by Lazaro Janier Gonzalez-Soler
online Big Blue Button Room: D19/2.03a, July 21, 2022 (Thursday), 12.00 noon
Keywords — Biometrics, Presentation Attack Detection, Face, Facial Regions, Facial Region Utility, Blending Ensemble
Nowadays, with the global spread of the SARS-CoV-2 coronavirus, wearing a mask has become one of the most efficient ways to prevent an infection. In addition, the use of accessories such as glasses or scarves, which occlude parts of the face, is also frequent in our daily life. Therefore, presentation attack detection (PAD) algorithms must efficiently detect differences between a bona fide presentation including the aforementioned occlusions and an intentional attack presentation. Since most PAD approaches focus on analysing the whole face, in this presentation we show the analysis of the feasibility of using different facial regions to improve the detection performance. Specifically, we define and investigate 14 facial regions such as the jaw, mouth, nose, and eyes. Furthermore, we propose a new metric, Face Region Utility, to determine those regions with high utility for detecting attack attempts.