FSAPF: A De-Scattering Framework With Stepwise Adjustment of Polarization Features.

Area of Science:

• Optics and Photonics
• Computer Vision
• Deep Learning

Background:

• Deep learning advances polarization imaging through scattering media.
• Existing methods struggle to effectively analyze and control polarization information during training.

Purpose of the Study:

• To propose a de-scattering framework with stepwise adjustment of polarization features (FSAPF) for high-performance imaging through scattering media.
• To enhance the analysis and control of polarization information in deep learning training.

Main Methods:

• Physically guided hierarchical learning, progressing from global structure to local polarization details.
• Introduction of a polarization learning module (PLM) to embed polarization priors and enforce physical consistency.
• Dynamic loss mechanism to enhance polarization features during training for improved robustness.

Main Results:

• The FSAPF framework demonstrates significant performance in target recovery tasks under various scattering conditions.
• Experimental validation confirms the superiority of FSAPF compared to existing methods.
• Ablation studies further support the effectiveness of the proposed components.

Conclusions:

• The proposed FSAPF framework offers a robust solution for high-performance polarization imaging through scattering media.
• The integration of polarization priors and physically guided learning enhances image quality and generalized robustness.
• The FSAPF framework shows significant potential for applications requiring clear imaging in turbid environments.