FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models

Authors: Zhipei Xu, Xuanyu Zhang, Runyi Li, Zecheng Tang, Qing Huang, Jian Zhang

Published: 2024-10-03 17:59:34+00:00

Comment: Accepted by ICLR 2025

AI Summary

FakeShield proposes an explainable Image Forgery Detection and Localization (IFDL) framework to address the black-box nature and limited generalization of existing methods. This multi-modal framework evaluates image authenticity, generates tampered region masks, and provides judgment bases using pixel-level and image-level clues. It leverages GPT-4o to enhance existing datasets into MMTD-Set and integrates a Domain Tag-guided Explainable Forgery Detection Module (DTE-FDM) with a Multi-modal Forgery Localization Module (MFLM).

Abstract

The rapid development of generative AI is a double-edged sword, which not only facilitates content creation but also makes image manipulation easier and more difficult to detect. Although current image forgery detection and localization (IFDL) methods are generally effective, they tend to face two challenges: \\textbf{1)} black-box nature with unknown detection principle, \\textbf{2)} limited generalization across diverse tampering methods (e.g., Photoshop, DeepFake, AIGC-Editing). To address these issues, we propose the explainable IFDL task and design FakeShield, a multi-modal framework capable of evaluating image authenticity, generating tampered region masks, and providing a judgment basis based on pixel-level and image-level tampering clues. Additionally, we leverage GPT-4o to enhance existing IFDL datasets, creating the Multi-Modal Tamper Description dataSet (MMTD-Set) for training FakeShield's tampering analysis capabilities. Meanwhile, we incorporate a Domain Tag-guided Explainable Forgery Detection Module (DTE-FDM) and a Multi-modal Forgery Localization Module (MFLM) to address various types of tamper detection interpretation and achieve forgery localization guided by detailed textual descriptions. Extensive experiments demonstrate that FakeShield effectively detects and localizes various tampering techniques, offering an explainable and superior solution compared to previous IFDL methods. The code is available at https://github.com/zhipeixu/FakeShield.

Key findings

FakeShield demonstrates superior performance in image forgery detection and localization across diverse tampering types, including Photoshop, DeepFake, and AIGC-Editing, achieving high accuracy and F1 scores. The framework effectively generates explainable judgments based on pixel-level and image-level anomalies, outperforming other M-LLMs in explanation quality. Additionally, it shows strong generalization and robustness to common image degradations like JPEG compression and Gaussian noise.

Approach

FakeShield is a multi-modal framework featuring a Domain Tag-guided Explainable Forgery Detection Module (DTE-FDM) and a Multi-modal Forgery Localization Module (MFLM). DTE-FDM uses a learnable domain tag generator and a fine-tuned LLM (LLaVA-v1.5-13B) to detect tampering and generate textual explanations. MFLM then leverages these explanations and image features with a Segment Anything Model (SAM) to produce precise tampered region masks.

Datasets

MMTD-Set (constructed from CASIAv2, Fantastic Reality, FFHQ, FaceApp, COCO, self-constructed AIGC-Editing data), CASIA1+, Columbia, IMD2020, Coverage, DSO, Korus, Seq-DeepFake, DFFD, ControlNet Inpainting, SDXL Inpainting.

Model(s)

UNKNOWN

Author countries

China

← Previous