Entropy guided method improves AI generated image detection accuracy
EIB-Net: Entropy-Guided Information Bottleneck for Generalizable AI-Generated Image Detection
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Detecting images made by artificial intelligence is hard because they often don’t have obvious fake parts. The authors found that AI images usually have smooth or less detailed areas that can help spot them. They created a technique called EIB-Net that picks out these smooth parts using a new measurement and focuses on them to learn better clues for detection. Their method works well even with little training data and works on images from different AI sources.
What this means in practice
- •For social media platforms: Detect AI-generated images quickly and accurately using limited data to prevent misinformation spread.
- •For digital forensics teams: Identify synthetic images from diverse AI generators by focusing on low-texture regions to support authenticity verification.
- •For content moderation services: Improve detection of AI-created images in user uploads using entropy-guided patch analysis to enhance moderation efficiency.$Commercial implications: Enabled development of proprietary detection tools for AI-generated image identification in large-scale content moderation.
Authors
Zhida Zhang, Xinlei Ma, Jie Cao
Abstract
The proliferation of photorealistic AI-generated images demands robust detection methods that generalize across diverse generative models. While existing approaches target manipulation-based forgeries with local artifacts, generation-based images (e.g., from diffusion models) lack such traces, posing a fundamental challenge. We observe that generative models prioritize global semantics at the expense of local texture fidelity, making low-texture regions key indicators of synthetic origin. To exploit this, we propose EIB-Net, an Entropy-guided Information Bottleneck Network. EIB-Net introduces a novel Image Entropy (IE) metric to automatically select the most informative (lowest-entropy) patch, then processes it with a Variational Information Bottleneck (VIB) to learn compact, generalizable features. Extensive experiments on DIFF, DiffusionForensics, and GenImage benchmarks demonstrate state-of-the-art performance: EIB-Net achieves 85.7\% accuracy using only 2\% of training data, outperforming full-image baselines by over 15\%, and maintains robust cross-generator generalization (83.5\% average accuracy on GenImage). Furthermore, our entropy-guided patch selection (EGPL) consistently enhances diverse backbones (CNNs and Transformers), proving its practical value for data-efficient detection.