A Learnable Gradient operator for face presentation attack detection

Caixun Wang, Bingyao Yu, Jie Zhou

Abstract: Highlights•A learnable gradient operator (LGO) is developed for face presentation attack presentation.•LGO is the generalization of some gradient operators such as the Sobel, which can efficiently extract distinguished fine-grained information and be deployed to frame-based or video-based approaches.•An adaptive gradient loss is proposed to better optimize the network.