da/sec scientific talk on Biometrics

Topic: Foundation Models for Face Presentation Attack Detection

by Lazaro Janier Gonzalez-Soler
D19/2.03a (also online via the corresponding BBB room), April 16, 2026 (Thursday), 12.00 noon

Keywords — Biometrics, Foundation Models, Presentation Attack Detection

Abstract

„Face recognition systems have achieved remarkable progress in recent years; however, they remain vulnerable to attack presentations (AP), which are often easy to fabricate and allow adversaries to impersonate authorised users. Modern deep learning–based presentation attack detection (PAD) systems require extensive annotated data and still exhibit limited generalisability when confronted with previously unseen presentation attack instruments (PAIs) or new datasets. This presentation investigates the use of large-scale foundation models to improve the robustness and adaptability of face PAD and introduces MixFoundPAD, a framework that combines multiple foundation models through parameter-efficient finetuning techniques to learn complementary representations for PAD.“