IGFD-Net: Illumination-guided frequency decoupling for polarization image fusion.

Area of Science:

• Computer Vision
• Image Processing
• Optics

Background:

• Intensity images (S0) capture scene radiance, while Degree of Linear Polarization (DoLP) reveals material properties.
• Effective fusion of S0 and DoLP enhances visual perception but is challenging under complex illumination.
• Existing methods often fail to maintain photometric consistency and saliency during fusion, causing artifacts.

Purpose of the Study:

• To develop a novel fusion method that addresses the limitations of existing polarization image fusion techniques.
• To improve robustness and accuracy in visual perception, especially under complex lighting conditions.
• To enhance downstream tasks like target detection through improved image fusion.

Main Methods:

• Proposed an illumination-guided frequency-decoupled fusion network (IGFD-Net).
• Introduced an illumination-guided attention mechanism to dynamically weight S0 and DoLP based on illumination priors.
• Implemented a frequency-decoupled fusion module for visual naturalness and a structure-aware saliency injection loss for photometric consistency.

Main Results:

• The IGFD-Net achieved competitive performance in subjective and objective image quality assessments.
• The method demonstrated superior ability in maintaining photometric consistency and preserving structural details compared to existing approaches.
• Fusion results showed improved accuracy in downstream target detection tasks.

Conclusions:

• The proposed IGFD-Net effectively fuses intensity and polarization images, overcoming limitations of prior methods.
• The illumination-guided and frequency-decoupled approach ensures robust visual perception and naturalness.
• This method offers significant improvements for applications requiring accurate visual information extraction and analysis.