Paper Image

Detecting fake images using natural image statistics

Published on:

25 March 2024

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Ziyou Liang,

Run Wang,

Weifeng Liu,

Yuyang Zhang,

Wenyuan Yang,

Lina Wang,

Xingkai Wang


Key Details

Proposes detecting fakes via proximity to learned 'natural traces' of real images

Natural traces capture intrinsic statistics stable across real images

Avoids reliance on artifacts of specific generative models

Shows high accuracy on GANs, diffusion models, and multi-step fakes

Robust to common image transformations

AI generated summary

Detecting fake images using natural image statistics

This paper proposes detecting fake images based on comparing them to stable 'natural traces' learned from real images, instead of focusing on artifacts from specific generative models. They employ statistical properties consistently present in real images to train a model to distinguish real from fake. Evaluation shows high accuracy in detecting various state-of-the-art generative models, and robustness to transformations.

Answers from this paper


No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up