Automatic liveness detection for facial images

Liveness detection is an important component in any facial biometrie system. It helps confirming that a real live person is in front of the camera. Given the ubiquity of the high-resolution printers and phone/tablet displays, popular attacks usually involve face prints or video replay. Most existing solutions rely on texture and local shape analysis to detect printing artifacts and light reflections in input attack images. In this paper, we propose extracting three low-level descriptors from the input face image, followed by polynomial classification and score level fusion. We show through our experiments how the fusion of multiple features and scores produced higher classification accuracy compared to the existing individual feature systems. We report our results on three popular benchmark datasets.