Face anti-spoofing with cross-stage relation enhancement and spoof material perception

Daiyuan Li, Guo Chen, Xixian Wu, Zitong Yu, Mingkui Tan

Abstract: Highlights•We propose a cross-stage fusion scheme that leverages semantically rich high-stage features to query task-relevant information in low-stage features, facilitating a more efficient cross-stage feature fusion.•We design an auxiliary pixel-wise material classification task to enhance the subtle discriminative capabilities of the features.•To address the limitations of existing Near-Infrared Dataset in the diversity of identities acquisition environment and acquisition devices. We collect a comprehensive Near-Infrared face anti-spoofing Dataset, which incorporates six illumination conditions and comprises 380,000 images from 1,040 distinct identities.