Face Authenticity Detection

The work presents a passive face authenticity detection system designed for employee time registration system and the data presenting its performance. The system is implemented as a stacking committee of models. Each of the four models included in the committee recognizes whether the camera view shows the real face of a live person or whether the face is shown in a photo or video. These models include: a convolutional network and Bezel algorithm for detecting portable devices on which the face can be shown, face context analysis, and a convolutional network for image analysis. The outputs from these models are fed to the inputs of a neural network that makes the final decision.