Spoofing Face Identification Using Higher order Descriptors

This paper proposes a novel higher order descriptors as Local Derivative Pattern(LDP), for face liveliness detection. We uses LDP frame work as directional descriptors based on change in texture of face. The nth order LDP descriptors are proposed to encode (n − 1)th directional derivatives to identify the texture for different kinds of face attack. LDP captures more detail information compare to first order LBP approach. We use LDP descriptor of derivatives order n = (2, 3, 4) and histogram bins are computed to form directional features for both genuine and different kinds of attack faces. An Multi layer perceptron (MLP) as week learner used for identification of genuine and attack face. Performance is evaluated by using MS-MSU database which consists of real genuine face and different kinds of attack face.