URFusion: Unsupervised Unified Degradation-Robust Image Fusion Network

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Existing image fusion methods struggle with low-quality images and specific degradations.
• A unified approach is needed to handle diverse image degradations during fusion.

Purpose of the Study:

• To propose an unsupervised, unified, degradation-robust image fusion network (URFusion).
• To uniformly eliminate various types of degradations for high-quality fused images.

Main Methods:

• URFusion employs three modules: intrinsic content extraction, intrinsic content fusion, and appearance representation learning and assignment.
• It extracts degradation-free content features and uses feature-level fusion constraints.
• Learns appearance representations and assigns high-quality statistical representations to the fused result.

Main Results:

• URFusion effectively eliminates degradation residues without needing ground truth.
• Demonstrated superior fusion performance on multi-exposure and multi-modal image fusion tasks.
• Successfully suppressed multiple types of degradations in fused images.

Conclusions:

• URFusion offers a robust and unified solution for image fusion across various degradations.
• The network achieves high-quality fused images by addressing degradation challenges effectively.