Comparative Analysis of CNN and RNN Techniques for Deepfake Video Detection

In the recent years photo shopping has been replaced by deepfakes. Deepfakes use deep learning to generate realistic media of fake events. The media can be of any form like audio, video or images. These deepfakes can be created without much efforts due to the presence of online tools and apps. If used with a bad intent, deepfakes are capable of causing harm to the society. Spread of mass misinformation, cyber bullying, spreading propaganda through fake news are few such examples. The rapid progression in deepfake media generation means that it is important to develop tools to identify such media. In this paper, we propose two methods to detect deepfake videos. The first method uses Convolutional Neural Networks (CNNs) and the second method uses Recurrent Neural Networks (RNNs). We train both the models on videos collected from three different open source datasets and also compare the performance of both methods.