Face liveness detection with recaptured feature extraction

Face recognition systems can be tricked by photos or videos with virtual faces. It is crucial for a safe face recognition system to distinguish genuine user's faces (i.e., the first captured images of real scene) and spoof faces (i.e., recaptured images of photographs or videos). Existing face liveness methods often use single image feature to address face spoofing problems, which are not reliable and robust. In this paper, we analyze the differences between genuine face images and spoof images, and propose to extract three types of features, i.e., specular reflection ratio, Hue channel distribution and blurriness, to determine whether a face image is captured from genuine face or not. Experimental results on NUAA photograph imposter database show the competitive performance of our method comparing with several state-of-the-art methods.