DRDFNet: A Degradation-Aware Restoration and Detail-Preserving Fusion Network for Infrared and Visible Image

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Multi-source image fusion, particularly infrared-visible image fusion (IVIF), is crucial for enhancing scene perception in applications like autonomous driving and drone reconnaissance.
• Existing IVIF methods often fail in adverse environments where visible images lose details and infrared images suffer from noise and artifacts, degrading fusion quality and downstream task performance.

Purpose of the Study:

• To propose a unified network, the Degradation-aware Restoration and Detail-preserving Fusion Network (DRDFNet), designed to address the limitations of current IVIF methods in adverse conditions.
• To improve the robustness and performance of image fusion for enhanced scene perception in challenging imaging scenarios.

Main Methods:

• DRDFNet integrates a Degradation-Aware Restoration Transformer with a Detail-Preserving Fusion Mamba.
• The restoration branch utilizes a Compound Degradation Restoration Module (CDRM) to mitigate complex image degradations.
• The fusion branch employs a Dynamic Feature Fusion Module (DFFM) for effective cross-modal feature integration, complemented by a two-stage training strategy to resolve optimization conflicts.

Main Results:

• Experiments conducted on the newly constructed DIVIF benchmark and the AWMM-100k dataset demonstrate DRDFNet's robust and competitive performance against state-of-the-art methods.
• The proposed DRDFNet effectively restores degraded images and preserves crucial details, leading to superior fusion quality.

Conclusions:

• DRDFNet offers a significant advancement in infrared-visible image fusion, particularly for applications operating in adverse environmental conditions.
• The developed dataset (DIVIF) and source code will facilitate further research and development in degraded image fusion.