Level Up the Deepfake Detection: a Method to Effectively Discriminate Images Generated by GAN Architectures and Diffusion Models

View on arXiv ← Back to list

Authors: Luca Guarnera, Oliver Giudice, Sebastiano Battiato

Published: 2023-03-01 16:01:46+00:00

AI Summary

This paper introduces a hierarchical multi-level approach for deepfake image detection and recognition, classifying images into 14 classes (9 GAN architectures, 4 diffusion models, and 1 real image class). The approach achieves over 97% accuracy in each classification task, outperforming state-of-the-art methods.

Abstract

The image deepfake detection task has been greatly addressed by the scientific community to discriminate real images from those generated by Artificial Intelligence (AI) models: a binary classification task. In this work, the deepfake detection and recognition task was investigated by collecting a dedicated dataset of pristine images and fake ones generated by 9 different Generative Adversarial Network (GAN) architectures and by 4 additional Diffusion Models (DM). A hierarchical multi-level approach was then introduced to solve three different deepfake detection and recognition tasks: (i) Real Vs AI generated; (ii) GANs Vs DMs; (iii) AI specific architecture recognition. Experimental results demonstrated, in each case, more than 97% classification accuracy, outperforming state-of-the-art methods.

Key findings

The hierarchical approach achieved over 97% accuracy in each classification level (real vs. AI, GAN vs. diffusion model, and specific architecture identification). This outperforms existing state-of-the-art methods, particularly in distinguishing between different diffusion models.

Approach

A hierarchical three-level classification approach is used. Level 1 distinguishes real from AI-generated images. Level 2 separates GAN-generated from diffusion model-generated images. Level 3 identifies the specific GAN or diffusion model architecture. ResNet-34 is used as the base model for each level.

Datasets

A custom dataset combining real images from CelebA, FFHQ, and ImageNet, and synthetic images generated by 9 GAN architectures (AttGAN, CycleGAN, GDWCT, IMLE, ProGAN, StarGAN, StarGAN-v2, StyleGAN, StyleGAN2) and 4 diffusion models (DALL-E 2, GLIDE, Latent Diffusion).

Model(s)

ResNet-34

Author countries

Italy

← Previous