Core Concepts
AI-generated images pose security risks, requiring improved detection methods. APN enhances cross-domain detection by extracting artifact features through explicit and implicit purification processes.
Abstract
The article introduces the Artifact Purification Network (APN) for detecting AI-generated images.
APN addresses performance drops in existing methods when faced with out-of-domain generators and scenes.
Explicit purification separates artifact features using frequency-band proposal and spatial decomposition methods.
Implicit purification further refines artifact-related features through mutual information estimation.
Experiments show APN outperforms previous methods in cross-generator and cross-scene detection.
Visualization analysis demonstrates the effectiveness of APN in extracting forgery patterns and condensing relevant information.
Stats
Experiments show that for cross-generator detection, the average accuracy of APN is 5.6% ∼ 16.4% higher than the previous 10 methods on GenImage dataset and 1.7% ∼ 50.1% on DiffusionForensics dataset.