Decoupling Semantics and Fingerprints: A Universal Representation for AI-Generated Image Detection

2026-05-07

AI Summary

Existing methods for detecting AI-generated images suffer from limited generalization, primarily due to the conflation of universal forgery traces, generator-specific fingerprints, and semantic content. This work reveals that fingerprints from different generators occupy non-overlapping subspaces in the frequency domain and introduces the Orthogonal Decomposition and Purification Network (ODP-Net), which leverages the physical orthogonality in the frequency domain to achieve the first structured disentanglement of these three components. By integrating instance-aware orthogonal decomposition, perturbation-driven purification, and manifold alignment, ODP-Net effectively separates feature subspaces, enhances semantic invariance, and promotes cross-domain alignment. The method achieves state-of-the-art performance on unseen generative models such as Stable Diffusion 3, demonstrating that disentangled representations are crucial for generalizable detection.

Abstract

Detecting AI-generated images across unseen architectures remains challenging, as existing models often overfit to generator-specific fingerprints and semantic content rather than learning universal forgery traces. We attribute this failure to feature entanglement: detectors learn these factors as a single entangled representation, where universal forgery traces are inextricably confounded with both generator-specific fingerprints and semantic content. Crucially, our spectral analysis reveals that this entanglement is avoidable: distinct generator-specific fingerprints (e.g., GAN stripes vs. Diffusion Model spots) occupy disjoint frequency subspaces and coexist as independent superpositions. Leveraging this physical orthogonality, we propose the Orthogonal Decomposition and Purification Network (ODP-Net) to structurally disentangle these factors. Specifically, ODP-Net employs (1) Instance-aware Orthogonal Decomposition to project features into mutually exclusive subspaces: universal forgery traces, generator-specific fingerprints, and semantic content; (2) Perturbation-based Purification to enforce semantic invariance via cross-sample feature injection; and (3) Manifold Alignment to bridge domain gaps. By explicitly decoupling universal forgery traces from generator-specific fingerprints and semantic content, ODP-Net achieves state-of-the-art performance on unseen architectures (e.g., Stable Diffusion 3), validating that structural disentanglement is key to generalization.

Problem

• AI-generated image detection
• generator-specific fingerprints
• semantic content
• universal forgery traces
• unseen architectures

Innovation

• orthogonal decomposition
• feature disentanglement
• AI-generated image detection
• frequency subspace
• universal forgery traces