Is JPEG AI going to change image forensics?

View on arXiv ← Back to list

Authors: Edoardo Daniele Cannas, Sara Mandelli, Nataša Popović, Ayman Alkhateeb, Alessandro Gnutti, Paolo Bestagini, Stefano Tubaro

Published: 2024-12-04 12:07:20+00:00

AI Summary

This paper investigates the counter-forensic effects of the JPEG AI compression standard on deepfake image detection and image splicing localization. Experiments show that JPEG AI reduces the performance of state-of-the-art forensic detectors by introducing artifacts similar to those found in manipulated images, highlighting the need for improved forensic techniques.

Abstract

In this paper, we investigate the counter-forensic effects of the new JPEG AI standard based on neural image compression, focusing on two critical areas: deepfake image detection and image splicing localization. Neural image compression leverages advanced neural network algorithms to achieve higher compression rates while maintaining image quality. However, it introduces artifacts that closely resemble those generated by image synthesis techniques and image splicing pipelines, complicating the work of researchers when discriminating pristine from manipulated content. We comprehensively analyze JPEG AI's counter-forensic effects through extensive experiments on several state-of-the-art detectors and datasets. Our results demonstrate a reduction in the performance of leading forensic detectors when analyzing content processed through JPEG AI. By exposing the vulnerabilities of the available forensic tools, we aim to raise the urgent need for multimedia forensics researchers to include JPEG AI images in their experimental setups and develop robust forensic techniques to distinguish between neural compression artifacts and actual manipulations.

Key findings

JPEG AI compression significantly reduces the accuracy of deepfake detectors, leading to an increase in false positives. Image splicing localization algorithms also suffer performance degradation due to JPEG AI artifacts. Retraining detectors with JPEG AI images shows some improvement but does not fully mitigate the problem.

Approach

The authors compressed images from various datasets using the official JPEG AI Reference Software at different bitrates (BPP). They then evaluated the performance of state-of-the-art deepfake detection and image splicing localization algorithms on these compressed images, comparing results to uncompressed images and those compressed using standard JPEG.

Datasets

CelebA, COCO, FFHQ, Imagenet, LAION, LSUN, RAISE, CASIA, Columbia, Coverage, COCOGlide, DSO-1

Model(s)

Several state-of-the-art deepfake detectors ([88]-A, [88]-B, [42], [32], [70], [34]-A, [34]-B, [81], [63]) and image splicing localization algorithms (TruFor, MMFusion, ImageForensicsOSN) were used.

Author countries

Italy

← Previous