Enhancing Video Forensics: Deep Learning Approaches to Combat Advanced Video Manipulation Techniques

Abstract

This research presents a method to efficiently detect facial tampering in videos, and particularly focuses on two recent techniques, used to generate hyper realistic forged videos: Deepfake and Face2Face. Traditional image forensics techniques are usually not well suited to videos due to the compression that strongly degrades the data. Thus, this work follows a deep learning approach and presents a network, with a smaller number of layers to focus on the mesoscopic properties of images. This work evaluates those fast networks on both an existing dataset and a dataset generated from online videos. Deepfake image detection is important because it helps everyone determine if the pictures seen online are real or fake. Due to advancements in computer vision techniques, people can create fake images that look extremely realistic. These could be used to spread lies or invade someone's privacy. The detection tools use smart technology to spot these fakes, ensuring that everyone can trust the pictures they come across and preventing the spread of misleading or harmful content on the internet. This work contributes to the growing body of research addressing the challenges posed by advanced video manipulation techniques, providing a valuable tool for applications in cybersecurity, media integrity, and the prevention of misinformation in an era dominated by sophisticated visual content manipulation.