Related Experiment Video
Updated: May 24, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
NSFF: Noise and semantic features fusion for AI-generated image detection
Haoran Yang1, Ruiqiang Ma1, Gang Wang1
1College of Intelligent Science and Technology, Inner Mongolia University of Technology, Hohhot, 010080, Inner Mongolia, China.
Summary
This study introduces NSFF, a novel framework for detecting AI-generated images by analyzing noise artifacts and semantic information. NSFF effectively identifies synthetic images from unseen models, improving misinformation detection.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Digital Forensics
Background:
- Advanced generative models create realistic AI-generated images, posing risks for misinformation spread.
- Existing AI-generated image detection methods require large datasets and struggle with novel generative models.
Purpose of the Study:
- To develop a general and robust detection framework for AI-generated images.
- To improve the generalization capability of AI-generated image detectors to unseen models.
Main Methods:
- Proposed NSFF (Noise and Semantic Fusion Framework) for AI-generated image detection.
- Utilized noise artifacts in poorly textured regions and combined global/local semantic information.
- Trained on a small dataset (2000 images) for supervised learning.
Main Results:
- NSFF demonstrated effective generalization to images from previously unseen generative models.
- Achieved an average accuracy of 80.82% on the test set.
- Outperformed previous methods relying solely on texture and semantic features.
Conclusions:
- NSFF offers a more robust and generalizable approach to detecting AI-generated images.
- The framework effectively addresses limitations of previous detection methods, particularly with novel generative models.
- This research contributes to combating the malicious use of AI-generated imagery.