Synthetic shadows: the interplay of forensic detection and anti-forensic techniques in GAN-generated images

Synthetic shadows: the interplay of forensic detection and anti-forensic techniques in GAN-generated images

연구 분야: Strategies

학회: Multimedia Systems

Generative adversarial networks (GANs) have become a leading innovation in the synthesis of realistic face images. These images are typically unrecognizable to the naked eye and find application in entertainment, virtual reality, and media, greatly challenging digital forensics. Photorealistic yet fake face images generated by GANs raise questions concerning their use in identity impersonation, disinformation attacks, and other nefarious activities in society. Anti-forensic techniques, whose purpose is to conceal the digital trail of synthetic media, introduce additional challenges to detecting images generated by GANs. These include techniques such as adding noise, adversarial perturbations, and compression artifacts that can be used for concealing from cybersecurity and law enforcement solutions. This survey discusses the mechanisms and challenges of face image detection of GAN-generated images, highlighting the emergent interplay between evolving forensic detection and counteracting anti-forensic tactics. Current detection methods—ranging from machine learning classification to deep learning models and frequency domain-based approaches—are promising but generally insufficient owing to dependence on specific training sets and the versatility of state-of-the-art GAN models. This review discusses anti-forensic techniques, such as adaptive GANs and adversarial attacks, which function to reduce the detectability of synthetic media. Comparisons of detection methods versus anti-forensic techniques reveal critical gaps in robustness and generalizability and necessitate interdisciplinary investigations to confront the evolving threat landscape. Future studies will revolve around the explainability and adaptability of the frameworks to detect sophisticated tactics related to anti-forensic techniques.