Adaptive learning-driven polarimetric dehazing imaging in a dynamic turbid environment.

Area of Science:

• Computer Vision
• Optical Engineering
• Image Processing

Background:

• Polarization dehazing imaging is crucial for enhancing image quality by mitigating scattering effects in complex environments.
• Existing methods primarily focus on homogeneous scattering, leaving dynamic, inhomogeneous scenes as a significant challenge.
• Underwater environments present unique difficulties due to suspended particles and dynamic scattering.

Purpose of the Study:

• To develop an effective method for dehazing images in dynamic, inhomogeneous underwater environments.
• To propose an adaptive learning-driven polarimetric dehazing imaging network (ALPD-Net).
• To improve image restoration performance in turbid underwater conditions.

Main Methods:

• An adaptive learning-driven polarimetric dehazing imaging network (ALPD-Net) was developed.
• The network accurately estimates two polarization-related parameters per pixel based on a physical dehazing model.
• Sequential multi-frame polarization images are fused to extract enhanced feature information for image recovery.

Main Results:

• The proposed ALPD-Net demonstrated superior convergence and restoration performance.
• Accurate estimation of polarization parameters across the scene was achieved.
• Experimental results confirmed the method's effectiveness and robustness compared to existing dehazing models.

Conclusions:

• The ALPD-Net effectively reduces inhomogeneous scattering effects in dynamic turbid underwater environments.
• The integration of polarization parameters and multi-frame fusion significantly improves image recovery.
• The proposed method offers a robust solution for practical applications requiring high-quality underwater imaging.