Revisiting Reconstruction-based AI-generated Image Detection: A Geometric Perspective

2025-10-28

AI Summary

Reconstruction-error-based methods for detecting AI-generated images lack theoretical foundations and often fail due to low reconstruction errors on authentic images, leading to poor separability. Method: This paper proposes ReGap, a training-free dynamic error analysis framework grounded in manifold geometry. Contribution/Results: First, it establishes a theoretical lower bound on the Jacobian spectrum, providing the first rigorous explanation of the intrinsic difference in reconstruction errors between authentic and generated images. Second, it introduces structured image editing as a controlled perturbation to define and compute the dynamic reconstruction error gap (ReGap) before and after editing—significantly enhancing error discriminability. Extensive experiments demonstrate that ReGap consistently outperforms state-of-the-art baselines across diverse generative models, post-processing operations, and adversarial corruptions, exhibiting strong generalization, robustness, and interpretability, with substantial improvements in detection accuracy.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

The rise of generative Artificial Intelligence (AI) has made detecting AI-generated images a critical challenge for ensuring authenticity. Existing reconstruction-based methods lack theoretical foundations and on empirical heuristics, limiting interpretability and reliability. In this paper, we introduce the Jacobian-Spectral Lower Bound for reconstruction error from a geometric perspective, showing that real images off the reconstruction manifold exhibit a non-trivial error lower bound, while generated images on the manifold have near-zero error. Furthermore, we reveal the limitations of existing methods that rely on static reconstruction error from a single pass. These methods often fail when some real images exhibit lower error than generated ones. This counterintuitive behavior reduces detection accuracy and requires data-specific threshold tuning, limiting their applicability in real-world scenarios. To address these challenges, we propose ReGap, a training-free method that computes dynamic reconstruction error by leveraging structured editing operations to introduce controlled perturbations. This enables measuring error changes before and after editing, improving detection accuracy by enhancing error separation. Experimental results show that our method outperforms existing baselines, exhibits robustness to common post-processing operations and generalizes effectively across diverse conditions.

Problem

• Detecting AI-generated images lacks theoretical foundations
• Static reconstruction error methods reduce detection accuracy
• Existing approaches require data-specific threshold tuning limitations

Innovation

• Introducing Jacobian-Spectral Lower Bound for reconstruction error
• Proposing ReGap method with dynamic reconstruction error computation
• Using structured editing operations to introduce controlled perturbations