UMCFuse: A Unified Multiple Complex Scenes Infrared and Visible Image Fusion Framework

Area of Science:

• Computer Vision
• Image Processing

Background:

• Infrared and visible image fusion is crucial but underperforms in complex scenes with interference.
• Existing methods struggle with detail preservation and noise reduction in challenging environments.

Purpose of the Study:

• To develop a unified framework, UMCFuse, for robust infrared and visible image fusion in complex scenes.
• To improve fusion performance by balancing interference removal and detail preservation.

Main Methods:

• Classifying visible image pixels based on light scattering to separate details from intensity.
• Implementing an adaptive denoising strategy for detail layer fusion.
• Fusing multi-modal energy features by analyzing them from multiple directions.

Main Results:

• UMCFuse demonstrates superior performance over representative methods on complex scene datasets.
• The method excels in adverse conditions like noise, blur, and overexposure.
• Effective fusion is shown across various downstream tasks, including segmentation and object detection.

Conclusions:

• UMCFuse provides a significant advancement in infrared and visible image fusion for complex scenes.
• The proposed adaptive denoising and multi-directional feature fusion strategies are key to its success.
• The framework generalizes well to diverse challenging scenarios and downstream applications.