Maximize margins for robust splicing detection

2025-07-28

AI Summary

Current image splicing detection models exhibit poor generalization against post-processing operations (e.g., JPEG compression, Gaussian filtering), severely undermining their reliability in real-world deployment. To address this, we propose a robust training paradigm grounded in latent-space decision boundary analysis: model robustness is quantified via boundary width, and models are jointly trained under multiple post-processing perturbations; the optimal checkpoint is selected on the validation set based on maximal boundary width. Crucially, this approach requires no architectural modifications or loss-function redesign—robustness is enhanced solely through refined training strategies and boundary-aware model selection, thereby improving discriminability in the feature space. Extensive experiments across multiple benchmark datasets demonstrate that our method significantly enhances resilience to common post-processing artifacts. Specifically, it yields average AUC improvements of 3.2–5.8 percentage points over state-of-the-art training strategies under JPEG compression and Gaussian blur.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

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

Abstract

Despite recent progress in splicing detection, deep learning-based forensic tools remain difficult to deploy in practice due to their high sensitivity to training conditions. Even mild post-processing applied to evaluation images can significantly degrade detector performance, raising concerns about their reliability in operational contexts. In this work, we show that the same deep architecture can react very differently to unseen post-processing depending on the learned weights, despite achieving similar accuracy on in-distribution test data. This variability stems from differences in the latent spaces induced by training, which affect how samples are separated internally. Our experiments reveal a strong correlation between the distribution of latent margins and a detector's ability to generalize to post-processed images. Based on this observation, we propose a practical strategy for building more robust detectors: train several variants of the same model under different conditions, and select the one that maximizes latent margins.

Problem

• Improve robustness of splicing detection against post-processing
• Address variability in deep learning model performance
• Maximize latent margins for better generalization

Innovation

• Maximize latent margins for robustness
• Train model variants under different conditions
• Select model with best generalization ability