An In-depth Review of Deepfake Detection Performance: Leveraging MesoNet, XceptionNet, and ResNet for Improved Accuracy and Robustness

Abstract

Over the past few years, artificial intelligence, machine learning and deep learning technologies have made substantial progress. Due to these developments, new techniques for handling multimedia, including images, videos, and audio, are now available. Technologies like these are mainly used to provide entertainment, to help teach and study and for research, though they have sometimes been exploited for negative reasons. One way this has been abused is by creating "Deepfakes." Deepfakes allow people in videos, images and sounds to appear as though they said or did what was constructed by the people making them. Many people have used this technology to spread misleading messages, start political disputes, and harm individuals by creating made-up content for blackmail or harassment. Experts have developed methods to detect these types of fake media. The authors of this paper provide a thorough review of several prior works on Deepfake detection.  It is found that Deepfakes constitute a significant problem, yet studies and new developments in AI and machine learning are helping develop better ways to identify them. Deep learning appears to be the most effective technique for detecting Deepfakes across a wide range of datasets [9]. The abstract covers how Deepfakes are becoming a challenge and the ways experts are working to uncover them. The paper demonstrates how features of deep learning strengthen its approaches and support their use in addressing the problems posed by Deepfakes.