Single Patch Based 3D High-Fidelity Mask Face Anti-Spoofing

Abstract

Face anti-spoofing is rapidly increasing in importance as facial recognition systems have become common in the financial and security fields. Among all kinds of attack, 3D high-fidelity masks are especially hard to defend. Recently, CASIA introduced a large scale dataset CASIA-SURF HiFiMask, which comprises of 54,600 videos recorded from 75 subjects with 225 high-fidelity masks. In this paper, we design a lightweight network with single patch input on the basis of CDCN++, and supervise it by focal loss. The proposed method achieves the Average Classification Error Rate (ACER) of 3.215 on the Protocol 3 of CASIASURF HiFiMask dataset and ranks the third best model in the Chalearn 3D High-Fidelity Mask Face Presentation Attack Detection Challenge at ICCV 2021.