IEEE Access
The paper entitled “A Novel Adversarial Gray-Box Attack on DCT-based Face Deepfake Detectors” has been accepted at IEEE Access 
This work introduces a novel adversarial gray-box attack specifically designed to bypass deepfake detectors based on Discrete Cosine Transform (DCT) features. The proposed method manipulates the AC coefficient statistics of synthetic images so that they closely resemble those of real ones, while preserving high visual quality. The attack assumes full knowledge of the DCT feature extraction process but no access to the internal parameters of the classifiers, making it a gray-box scenario. Experimental results show that the attack is effective against detectors tested on deepfakes generated by both GANs and Diffusion Models, significantly reducing their reliability and exposing critical vulnerabilities in detection systems often considered interpretable and robust. This research highlights important implications for multimedia forensics and cybersecurity, raising awareness of the need for more resilient defense strategies against adversarial manipulations.
Authors: Francesco Guarnera, Luca Guarnera, Alessandro Ortis, Sebastiano Battiato, Giovanni Puglisi.Link: https://ieeexplore.ieee.org/document/11192411
