Facial Behavior Analysis-Based Deepfake Video Detection using GAN Discriminator

Abstract

Deepfake is an artificial intelligence-based method for making fake images of people. It works by putting the existing (source) images or videos on the final (destination) images or videos. But recent improvements in deep learning have made it much easier to make fake videos that look real and are convincing with a relatively small amount of data and computing power. As a result of the development of deep learning techniques such as Generative adversarial networks (GAN), Deepfake has become closer to the truth. Many researchers are based on discovering deep fakes that were created by traditional methods. Traditional methods of detection that look for artifacts and pixels that don't match up can't keep up. This paper can detect deepfakes that are perfectly created. It is detected by modifying the GAN algorithm and inverting its function. The discriminator model of a GAN network is used to analyze behavior, facial gestures, and the appearance of an object. The paper is divided into two stages. The first stage is to use a GAN discriminator that has been modified. It is then trained using a deepfake dataset. The second stage is to test the videos by extracting the faces. Next, run it through the GAN discriminator to see if it's a forgery. In comparison to other networks, the GAN discriminator has demonstrated its ability and accuracy in detecting fake videos. The network's accuracy in detecting and distinguishing between real and fake videos is %94.65.