Face Anti-Spoofing by Spatial Fusion of Colour Texture Features and Deep Features
Arnav Anand, Dinesh Kumar Vishwakarma · 2020
Biological features based identification has seen a surge in recent times. Nonetheless, a number of spoofing skills have played a negative role in the ever increasing affluence of innovation in bio-metrics technology, specifically in the areas of face detection and face recognition. As a solution for the above mentioned issue, more vigorous as well as precise face antispoofing strategies have been developed. Deep Learning with Convolutional Neural Nets (CNNs) have shown exceptional success where liveness detection of the face is concerned. In this thesis, a constructive approach has been proposed which would detect Face Anti-Spoofing with the help of CNNs and Local Binary Patterns (or LBPs). Firstly, CNNs are used to extract the global/deep features followed by LBPs to extract the local/color texture features. After this, the two different kinds of features predict whether a face is genuine or spoofed with use of Support Vector Machines (SVMs) for the classification and then two different lists of probabilities are generated from both the models with different features. These probabilities are then fused together to recognize non-spoofed faces from spoofed faces.