A Novel Methodology for Deepfake Detection Using MesoNet and GAN-based Deepfake Creation
Aung Win, Myo Min Hein, Chit Htay Lwin, Aung Myo Thu, Myo Myat Thu, Nu Yin Khaing · 2024
In the past few years, the development of "Deepfake" videos, which are defined by the use of free software tools based on deep learning, has enabled the production of credible face exchanges in videos with minimum evidence of manipulation. Artificial intelligence-altered videos which have gained widespread attention are known as "Deepfake" videos. They are employed to highly realistically manipulate content (audio, video, and image) to create an appearance that a person said or did something they never actually did. The purpose of these deepfake videos is to spread disinformation about people, including governments and personalities, through the use of deep learning algorithms. These films have been intentionally made global with the aim of frightening people, spreading misinformation and marketing, and disrupting society. Deepfake technology uses algorithms like Generative Adversarial Network (GAN) to produce artificial images or videos that look realistic. In GANs, a generator and discriminator collaborate to produce phony images or videos that have an authentic look. DeepFaceLab is a free tool for creating deepfakes with a lot of generating flexibility. A deepfake video is quite hard for an average viewer to identify. In order to improve the performance of deepfake detection, we proposed a novel MesoNet approach in this research study that can effectively identify photographs as real or fake. The results of the experiment showed that the enhanced classifier (Meso) network model performed about 90% in terms of identifying real from false photographs.