Abstract: Over the past few decades, there have been rapid breakthroughs in AI, machine learning, and deep learning, which have led to the development of new tools and methodologies for manipulating multimedia. Technology has generally been employed for beneficial purposes, such as education and entertainment, but dishonest users have also exploited it for darker or illicit purposes. For example, incredibly realistic-looking, well-produced fake movies, images, or audio have been made to spread false information, incite political unrest and hatred, or even to harass and blackmail people. The extremely replicated, realistic, and edited videos have been dubbed as "Deepfake" in recent times. Since then, several approaches to resolving the problems raised by Deepfake have been described in length in the literature. We conduct a systematic literature review (SLR) in this work to give a current synopsis of the Deepfake detection research projects. We provide an overview of 112 pertinent publications from 2018 to 2020 that showcased various methodologies. For analysis, we divide them into four categories: deep learning-based approaches, traditional machine learning-based approaches, statistical-based approaches, and blockchain-based approaches. We also evaluate the pattern recognition performance of several algorithms on various datasets, and we find that deep learningbased approaches outperform other approaches in Deepfake detection
Read full abstract