On the Exploitation of DCT-Traces in the Generative-AI Domain

View on arXiv ← Back to list

Authors: Orazio Pontorno, Luca Guarnera, Sebastiano Battiato

Published: 2024-02-03 16:45:31+00:00

AI Summary

This paper investigates the use of Discrete Cosine Transform (DCT) coefficient statistics to detect deepfake images generated by GANs and Diffusion Models. By analyzing the statistical distribution of these coefficients, particularly the βAC values, the authors identify discriminative fingerprints that persist even after JPEG compression.

Abstract

Deepfakes represent one of the toughest challenges in the world of Cybersecurity and Digital Forensics, especially considering the high-quality results obtained with recent generative AI-based solutions. Almost all generative models leave unique traces in synthetic data that, if analyzed and identified in detail, can be exploited to improve the generalization limitations of existing deepfake detectors. In this paper we analyzed deepfake images in the frequency domain generated by both GAN and Diffusion Model engines, examining in detail the underlying statistical distribution of Discrete Cosine Transform (DCT) coefficients. Recognizing that not all coefficients contribute equally to image detection, we hypothesize the existence of a unique ``discriminative fingerprint, embedded in specific combinations of coefficients. To identify them, Machine Learning classifiers were trained on various combinations of coefficients. In addition, the Explainable AI (XAI) LIME algorithm was used to search for intrinsic discriminative combinations of coefficients. Finally, we performed a robustness test to analyze the persistence of traces by applying JPEG compression. The experimental results reveal the existence of traces left by the generative models that are more discriminative and persistent at JPEG attacks. Code and dataset are available at https://github.com/opontorno/dcts_analysis_deepfakes.

Key findings

The study reveals discriminative traces in the DCT coefficients, particularly in high-frequency components for raw images. While these traces fade with JPEG compression, subsets identified using LIME show better robustness. Low-frequency coefficients show more resistance to compression attacks.

Approach

The authors analyze the statistical distribution of Discrete Cosine Transform (DCT) coefficients in deepfake images. They train machine learning classifiers on various combinations of these coefficients, using LIME for feature selection, to identify discriminative fingerprints and test their robustness to JPEG compression.

Datasets

A dataset of 72,334 images, including 19,334 real images and images generated by various GAN and Diffusion Model architectures (e.g., StyleGAN, Stable Diffusion, DALL-E 2). Sources include CelebA, FFHQ, and data generated by the authors.

Model(s)

K-Nearest Neighbors (K-NN), Gradient Boosting, Random Forest, and a custom neural network for LIME feature selection.

Author countries

Italy

← Previous