Face Spoof Detection Using Dual Stream Convolutional Neural Network

Abstract

Face recognition is an acknowledgement strategy used to distinguish countenances of people whose pictures spared in information index. Face acknowledgement has consistently stayed a huge focal point of research due to its non-intruding nature. Face discovery is utilized in biometrics, regularly as a piece of a facial acknowledgement framework. This imaginative innovation has defects. This is the place the requirement for against parodying arrangements becomes possibly the most important factor. Most of the faces caricaturing assaults utilize 2D and 3D to trick facial acknowledgement programming. In spite of the fact that numerous compelling strategies have been proposed for against parodying we find that the presentation of many existing techniques is corrupted by illumination. It spurs us to create light invariant strategies for hostile to satirizing. In our paper we propose a double stream convolutional neural system. It fundamentally chips away at two spaces: to be specific RGB and MSR. The two spaces are similarly significant on the grounds that the previous contains high recurrence facial highlights yet delicate to brightening. Both these highlights are taken care of the system and we use attention based combination strategy to intertwine both the features. The outcome, whether the picture is genuine or parody is built up utilizing softmax.

 Keywords: Face spoofing, convolutional neural networks, attention-based fusion, softmax.