Infrared and visible image fusion algorithm based on gradient attention residuals dense block.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Infrared and visible image fusion aims to combine thermal and visual data.
• Existing methods often neglect attention mechanisms, leading to loss of visible texture information.
• There is a need for fusion methods that preserve details from both modalities.

Purpose of the Study:

• To propose a new infrared and visible image fusion method.
• To enhance the extraction of important features and retention of image details.
• To improve the quality of fused images by preserving texture information.

Main Methods:

• Developed a novel gradient attention residual dense block integrating squeeze-and-excitation networks.
• Introduced a feature gradient attention module for improved detail retention.
• Employed an adaptive weighted energy attention network in the fusion layer for enhanced detail preservation.

Main Results:

• The proposed method demonstrated superior performance on the TNO dataset across multiple evaluation metrics.
• Achieved significant improvements in average gradient (AG), information entropy (EN), spatial frequency (SF), mutual information (MI), and standard deviation (SD).
• Outperformed five common fusion methods by notable margins in all tested indicators.

Conclusions:

• The proposed dense gradient attention residual method is effective for infrared and visible image fusion.
• The method excels at preserving both infrared targets and visible light texture information.
• The approach offers a superior alternative to existing fusion techniques, validated by extensive experimental results.