Oscillating patterns based face antispoofing approach against video replay

Typically, an automatic face authentication (FA) procedure begins with data (facial images) acquisition, procedure that can be carried out with or without human monitoring (in unconstrained settings), the subsequent steps being automatically processed. When the human monitoring is absent for the access procedure (i.e., the system is operating in the “wild”), the current FA systems can be easily cheated by spoofing identities using photographs or recorded video playback containing genuine information. The aim of this paper is to present an approach to indicating potential spoof attacks when a video recording of a genuine user is playback in front of a FA system. The approach relies on detecting specific image artifacts, more precisely oscillating patterns. Smooth image areas are first identified in the pixel domain as containing potential oscillating-like patterns. Several image statistics are next extracted and corresponding feature vectors are formed. Eventually, these feature vectors are classified as real or attack feature vectors by means of Lagrangian Support Vector Machines (LSVMs). When compared to two state-of-the-art methods, namely local binary patterns (LBP) and concentric Fourier based features, the experimental results indicate that the proposed approach substantially outperforms the two for this particular type of video data.