Towards Good Practices in Face Anti-Spoofing: An Image Reconstruction Based Method
Abstract
Face anti-spoofing is vital to the security of face recognition. Previous methods rely on hand-crafted features or use estimate depth map as supervision in CNNs and result in poor performance. In this paper, we propose a novel auto-encoder image reconstruction model to address the face anti-spoofing problem. In our model, we introduce the multi-scale skip connections to excavate multi-level semantics contained in images. Further, we combine the SSIM loss and Ll loss to guide the network towards better reconstruction results. We verify the effectiveness of our model design on OULU-NPU and SiW datasets and obtain promising results.