Infrared and Harsh Light Visible Image Fusion Using an Environmental Light Perception Network

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Standard image fusion algorithms struggle with harsh nighttime lighting, producing low-quality fused images.
• Visible light images in these conditions have low information entropy and high pixel intensity, creating a contradiction.
• This limits the effectiveness of fusion for high-level vision tasks like intelligent driving.

Purpose of the Study:

• To develop a robust image fusion network resilient to harsh light interference.
• To enhance information entropy and retain crucial details in fused infrared and visible images.
• To improve performance in nighttime intelligent driving applications.

Main Methods:

• Designed an edge feature extraction module to optimize fusion information entropy.
• Proposed a harsh light environment aware (HLEA) module to address image quality degradation.
• Developed an edge-guided hierarchical fusion (EGHF) module for robust feature fusion.

Main Results:

• The proposed method significantly enhances information entropy in fused images under harsh lighting.
• Fusion results contain more useful information compared to existing advanced algorithms.
• Demonstrated superior performance in high-level vision tasks under challenging nighttime conditions.

Conclusions:

• The novel image fusion network effectively overcomes limitations of existing methods in harsh nighttime environments.
• The approach provides substantial assistance for nighttime vehicle intelligent driving systems.
• The integration of entropy and information theory principles enhances fusion robustness and information retention.