Micro-Defects Expose Macro-Fakes: Detecting AI-Generated Images via Local Distributional Shifts

2026-05-09

AI Summary

Existing methods for detecting AI-generated images often rely heavily on global semantic features, making it difficult to effectively capture subtle, localized artifacts of forgery. To address this limitation, this work proposes the MDMF framework, which introduces a novel learnable Patch Forensic Signature that maps local image patches into a compact forensic embedding space. By modeling the statistical distribution of these local signatures, MDMF amplifies microscopic anomalies into macroscopic distributional discrepancies. Theoretical analysis demonstrates that this localized modeling strategy yields significantly enhanced discriminability between real and synthetic images. Leveraging the Maximum Mean Discrepancy (MMD) as a distributional metric, MDMF achieves state-of-the-art performance across multiple benchmark datasets, underscoring its robustness and generalizability in detecting AI-generated imagery.

Abstract

Recent generative models can produce images that appear highly realistic, raising challenges in distinguishing real and AI-generated images. Yet existing detectors based on pre-trained feature extractors tend to over-rely on global semantics, limiting sensitivity to the critical micro-defects. In this work, we propose Micro-Defects expose Macro-Fakes (MDMF), a local distribution-aware detection framework that amplifies micro-scale statistical irregularities into macro-level distributional discrepancies. To avoid localized forensic cues being diluted by plain aggregation, we introduce a learnable Patch Forensic Signature that projects semantic patch embeddings into a compact forensic latent space. We then use Maximum Mean Discrepancy (MMD) to quantify distributional discrepancies between generated and real images. Our theory-grounded analysis shows that patch-wise modeling yields provably larger discrepancies when localized forensic signals are present in generated images, enabling more reliable separation from real images. Extensive experiments demonstrate that MDMF consistently outperforms baseline detectors across multiple benchmarks, validating its general effectiveness. Project page: https://zbox1005.github.io/MDMF-project/

Problem

• AI-generated images
• micro-defects
• image forensics
• distributional shifts
• deepfake detection

Innovation

• micro-defects
• local distributional shifts
• Patch Forensic Signature
• Maximum Mean Discrepancy
• AI-generated image detection