A Dual-Modal Face Anti-Spoofing Method via Light-Weight Networks

Face anti-spoofing is very significant for the security of face recognition systems. In this paper, we utilize the near-infrared (NIR) and visible light (VIS) modalities to distinguish the living face from the presentation attacks. We explore different dual-modal combination strategies and propose a dual-modal light-weight network based method for face anti-spoofing. We modify the moblienetv3 and take it as the backbone network to extract the clues from NIR-VIS image pairs. Then we merge and select the feature in the embedding space before the classification. In extensive experiments on self-collected dataset, we demonstrate that the proposed method is efficient, effective and robust.