Anti-spoofing

An anti-spoofing system 602 is disclosed which comprises a depth estimation component, a global anti-spoofing classifier, and a patch-based anti-spoofing classifier. The depth estimation component receives a 2D verification image (206) and extracts estimated depth information therefrom. The global anti-spoofing classifier 504 uses the extracted depth information to classify the 2D verification image in relation to real (actual humans) and spoofing classes, and thereby assigns a global classification value to the whole of the image. The patch-based anti-spoofing classifier 1102a,b classifies each image patch of the 2D verification image in relation to the real and anti-spoofing classes, and thereby assigns a local classification value to each image patch. The system combines 1104 the global and local classification values to determine whether an entity captured in the 2D verification image corresponds to an actual human or a spoofing entity. The patched-based classifier could employ convolutional neural networks 110a,b to define patches. The methods could be used to detect mask, cut-out, replay or print attacks.