Year: 2025
Type: article
Abstract: The rise of deepfake media powered by the abrupt and rapid advances in generative AI, has introduced serious concerns and challenges in digital content verification. Although there is a wide range of detection strategies currently in use and being proposed, many remain limited in their ability to generalize across manipulation types, defend against adversarial attacks, or operate efficiently in real world scenarios. This paper presents a survey of deepfake detection and defense techniques, organized into ten technical categories: frequency-based, spatial-based, temporal-based, physiological-based, multimodal detection, adversarial attacks and defenses, forensic noise-based methods, watermarking and prevention strategies and dataset contributions. Drawing on one hundred studies, this study analyzes the design, strengths, and weaknesses of the proposed approaches in each category. Comparative tables and short summaries are also provided discussing how the models function, the performance trends, robustness under varying conditions, and the common trade-offs. A discussion then follows which identifies major research gaps such as limited dataset diversity, weak adversarial robustness, and high computational demands. The study then concludes by recommending directions for future research, including hybrid detection models, adaptive adversarial defenses, and the broader integration of multimodal and forensic techniques. This paper contributes to a better understanding of the current state and future direction of the deepfake detection and defense landscape. This insight is critical in ensuring the integrity of digital media in this era of rapidly advancing generative manipulation that shows no sign of slowing any time soon.
Author: Francis, Nigel
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7113001108
