An Effective Solution Towards Solving the Problem of Deepfake
Keywords:
Residual Network, Deepfake, LSTM, Inception ResNetAbstract
The use of advanced AI algorithms is growing daily, resulting in increasingly realistic fake videos with facial superimposition that are being produced by AI, as these videos involve well-known people, their behavior can have severe consequences. The videos have the potential to not only spread false information that harms the status of individuals, businesses, and nations, but also to trigger widespread hysteria and civil unrest. Therefore, to stop the rapid spread of Deepfake it is of extreme importance to identify it first. A novel Residual Network based model that learns inherently distinct patterns that change between a real video and a Deepfake is proposed in this paper as a method for detecting Deepfakes. An Inception Residual Network Version 2 has been trained by the proposed model to successfully differentiate Deepfake videos using different features of a frame-based approach. This work shows that our model can be used to accurately identify Deepfake faces in each video source. This will help security applications effectively reduce the ever-increasing threat of Deepfakes.