Vtd-Net: Depth Face Forgery Oriented Video Tampering Detection Based On Convolutional Neural Network

Face is the basis of identity authentication in many software, and the rise of generative adversary network makes the forgery of face easier than ever, which brings great challenges to information security. We propose a novel deep convolution neural network called VTD-Net to recognize faces generated by adversarial learning. The network is full-pipeline which composed of face location, interception, scaling and detection. In the experiment, we use the latest challenging face forgery dataset Celeb-DF, evaluated the forgery detection performance at frame level and video level, and achieved state-of-the-art results.