Deepfake Video Detection Based on Convolutional Neural Networks

The increasing use of mobile camera technology and the growth of social media using and sharing have made the generation and publishing of digital videos more suitable than ever before. However, the manipulation and fabrication of videos have decreased in recent years, because of machine learning and computer vision techniques. This study uses the detection method that applies by comparing the areas of the generated face and their surrounding areas with the Convolutional Neural Network (CNN) Model. The model was applied to the DFDC dataset with different 60 clips for real and fake. The methodology of this work passes through three stages, the first stage pass through preprocessing to convert each video into frames and detect the face in each frame and then be cropped by Haar Cascade function; in the feature's extraction stage the ResNet-50 is applied as the feature extraction model. In the last stage, the CNN classifier for detecting whether the image is fake or real. From the experiment, The Deepfake detection method detects the fake face in the video, where the detection accuracy was achieved at 98%.