AI Face-Swap Detection

Abstract

With the growth of face-swapping technologies enabled by deep learning and based on generative adversarial networks, concerns about the validity of digital media and its misuse to perpetrate misinformation, identity theft, and digital whittling have escalated. This paper provides a comprehensive review and taxonomy of recent developments in face-swap detection, including performance gaps at both the algorithmic and deployment levels. A reproducible search of Scopus, IEEE Xplore, ACM Digital Library, ScienceDirect, and SpringerLink was conducted for studies published from January 2021 to March 2025, yielding 47 eligible peer-reviewed studies, of which 40 were suitable for comparative synthesis. The review categorizes detection strategies into deep learning-based, shallow, hybrid, and handcrafted feature-based models and evaluates their architectures, datasets, and performance measures. The reviewed benchmark datasets include FaceForensics++, Celeb-DF, the DeepFake Detection Challenge, DeeperForensics-1.0, and WildDeepfake.