DeepFake Videos Detection by Using Recurrent Neural Network (RNN)
Ali Abdulzahra Mohsin Albazony, Haider A Al-Wzwazy, Ahmed Salih Al-Khaleefa, Murtadha A. Alazzawi, Mohammed Almohamadi, Seyed Enayatallah Alavi
Abstract
In the last few years, the increasing development of various tools to make fake videos from real videos has been raised. Thus, several models/approaches have been constructed to detect and reveals fake video. Consequently, this research is conducted to propose a new model based on combining Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and image preprocessing techniques to classify and find the fake video from the real video. To implement and evaluate the proposed model, a MATLAB simulator has been used. The deepFake Images dataset is used for evaluations. This dataset contains 135 real videos as well as 677 fake videos created using different tools on real videos. Two scenarios have been utilized to evaluate the performance of the proposed model which are; dimensions of the training data and different sizes of the training data. The results show that the proposed model has been able to provide better results than the previous model.