New method detects synthetic images using noise residuals and clustering

Researchers have developed a novel, training-free method for detecting synthetic images by analyzing noise residual fingerprints. This approach utilizes a pre-trained Noiseprint++ model to extract these fingerprints, followed by feature extraction using a frozen Vision Transformer (ViT) and adaptive weighted fusion. Unsupervised K-Means clustering is then employed with minimal real image samples to distinguish between real and synthetic images, achieving an average accuracy of 82.2% on benchmark datasets and demonstrating superior generalization capabilities, particularly against diffusion-generated images. AI

IMPACT This training-free detection method could improve the security and trustworthiness of digital images by providing a more accessible and generalizable way to identify AI-generated content.

RANK_REASON The item is an academic paper detailing a new method for synthetic image detection. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• k-means clustering
• Noiseprint++
• ScienceCast
• vision transformer

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →