ADCD-Net: Robust Document Image Forgery Localization via Adaptive DCT Feature and Hierarchical Content Disentanglement

2025-07-22

AI Summary

Document image forgery localization faces two key challenges: general natural-image methods lack sensitivity to subtle manipulations in structured text, while specialized approaches exhibit insufficient robustness against common degradations such as compression and blurring. To address these issues, we propose a robust document forgery localization framework. Our method introduces three core innovations: (1) an adaptive DCT feature modulation mechanism that dynamically weights frequency-domain features based on block-aligned scoring; (2) a hierarchical content disentanglement network that explicitly separates textual content, background, and tampering cues; and (3) a dual-stream RGB/DCT feature fusion architecture integrated with prototype-guided modeling of pristine regions and multi-scale contextual aggregation. Evaluated under five representative distortion types, our approach achieves a 20.79% average performance gain over state-of-the-art methods, significantly improving localization accuracy and generalization under complex degradation conditions.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Feature Construction/Reformulation
• Natural Language Processing: Safety and Robustness

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

The advancement of image editing tools has enabled malicious manipulation of sensitive document images, underscoring the need for robust document image forgery detection.Though forgery detectors for natural images have been extensively studied, they struggle with document images, as the tampered regions can be seamlessly blended into the uniform document background (BG) and structured text. On the other hand, existing document-specific methods lack sufficient robustness against various degradations, which limits their practical deployment. This paper presents ADCD-Net, a robust document forgery localization model that adaptively leverages the RGB/DCT forensic traces and integrates key characteristics of document images. Specifically, to address the DCT traces' sensitivity to block misalignment, we adaptively modulate the DCT feature contribution based on a predicted alignment score, resulting in much improved resilience to various distortions, including resizing and cropping. Also, a hierarchical content disentanglement approach is proposed to boost the localization performance via mitigating the text-BG disparities. Furthermore, noticing the predominantly pristine nature of BG regions, we construct a pristine prototype capturing traces of untampered regions, and eventually enhance both the localization accuracy and robustness. Our proposed ADCD-Net demonstrates superior forgery localization performance, consistently outperforming state-of-the-art methods by 20.79% averaged over 5 types of distortions. The code is available at https://github.com/KAHIMWONG/ACDC-Net.

Problem

• Detect forged regions in document images robustly
• Overcome sensitivity of DCT traces to distortions
• Improve localization by disentangling text and background

Innovation

• Adaptive DCT feature modulation for alignment resilience
• Hierarchical content disentanglement to reduce text-BG disparities
• Pristine prototype construction for enhanced localization accuracy