Deepfake detection: critical review of state-of-the-art approaches and future perspectives

Abstract In recent years, the advancement of technology, particularly in domains such as AI, machine learning, and deep learning, has fostered the development of novel tools for altering visual media, including images, audios, and videos. Among these tools, deepfakes have emerged as a prominent example, leveraging generative adversarial networks to autonomously create deceptive visual content. The proliferation of deepfake technology has raised significant concerns due to its potential for misuse, such as in elections and social media platforms. This review provides an in-depth exploration of deepfakes, including their generation process, various detection methodologies, and established performance benchmarks. We examine primary methods employed to manipulate facial features within deepfake content, encompassing complete facial synthesis, identity swapping, attribute alteration, and expression substitution. The review also discusses the challenges associated with deepfake detection, highlighting the continuous evolution of deepfake generation techniques and tools, making detection increasingly challenging. Our findings underscore the importance of ongoing research in developing robust deepfake detection methods to mitigate potential harms associated with the spread of deceptive visual content.