Deep and Wide Features for Face Anti-Spoofing

Abstract

Recently, the face recognition system is rapid development, and face anti-spoofing (FAS) also plays a significant role in that system. There are two approaches to FAS methods 1) use handcrafted features and 2) use deep features extracted from deep learning networks. In this paper, we propose an end-to-end framework that combines wide and deep features to detect real and spoof images in the FAS problem. We also evaluated the effectiveness of our methods on different FAS datasets such as CASIA-FASD, MSU-MFSD, and ROSE-YOUTU.