Presentation attack detection (PAD) is a critical component in secure face\nauthentication. We present a PAD algorithm to distinguish face spoofs generated\nby a photograph of a subject from live images. Our method uses an image\ndecomposition network to extract albedo and normal. The domain gap between the\nreal and spoof face images leads to easily identifiable differences, especially\nbetween the recovered albedo maps. We enhance this domain gap by retraining\nexisting methods using supervised contrastive loss. We present empirical and\ntheoretical analysis that demonstrates that contrast and lighting effects can\nplay a significant role in PAD; these show up, particularly in the recovered\nalbedo. Finally, we demonstrate that by combining all of these methods we\nachieve state-of-the-art results on both intra-dataset testing for\nCelebA-Spoof, OULU, CASIA-SURF datasets and inter-dataset setting on SiW,\nCASIA-MFSD, Replay-Attack and MSU-MFSD datasets.\n