Neural network method for detecting video images of a person in a video stream of the facial biometrics system

The problem of recognizing spoofing attacks on a facial biometrics system is considered. It is based on replacing the real person’s face with a video image of another person’s face formed on the device screen in the field of view of a video surveillance camera. A neural network method for detecting such substitutions is proposed, which uses the results of matrix preprocessing of a sequence of video frames as input. A comparative analysis of the preprocessing algorithms is carried out. Estimates of the prediction quality metrics for the selected network architecture are obtained.