Abstract

The deepfake technology is arguably one of the most alarming digital threats of the present world. With the help of AI and deep generative models, in a few minutes and by almost any person, a highly convincing fake video can be made without the need for advanced technical skills. Such videos have the potential to mislead the public, damage reputations, and manipulate people’s opinions massively. While deepfake videos are proliferating on social media as well as in the case of video-conferencing platforms such as YouTube, Zoom, and Google Meet, the need for real-time and precise detection methods has become indispensable. This article provides a comprehensive and human-centric review that focuses on deepfake video detection methods. It largely focuses on deep learning, conventional machine learning, computer vision, and hybrid system methods. Besides, the review paper proposes an extensive hybrid strategy that combines CNN, LSTM, and Transformer architectures with Big Data analytics, not only to enable more robust detection but also to allow scalable detection. Moreover, the paper discussion summarizes the recent accomplishments, recognizes the existing shortcomings, and proposes the follow-up research direction for detection systems that are interpretable, less computationally intensive, and privacy-friendly.