An end-to-end sea fog removal network using multiple scattering model.

Area of Science:

• Computer Vision
• Image Processing
• Atmospheric Optics

Background:

• Sea fog significantly degrades image quality, particularly in oceanic scenes.
• Existing single scattering models fail to adequately address the complexities of dense fog.
• Image blurring and artifacts are common challenges in fog removal algorithms.

Purpose of the Study:

• To propose an end-to-end deep learning network for effective sea fog removal.
• To leverage a multiple scattering model for more accurate fog density representation.
• To improve image quality by mitigating blurring and artifacts.

Main Methods:

• Developed an end-to-end network incorporating a re-formulated atmospheric multiple scattering model.
• Unified transmission map, atmospheric light, and blur kernel into a single formula.
• Employed smooth dilation, sub-pixel techniques, and a multi-scale sub-network.
• Integrated multiple loss functions for comprehensive network training.

Main Results:

• The proposed model effectively removes sea fog, producing clearer images.
• Quantitative and qualitative experimental results demonstrate superiority over state-of-the-art methods.
• The multiple scattering model significantly reduces image blurring compared to single scattering models.
• Advanced techniques successfully avoided gridding and halo artifacts.

Conclusions:

• The developed network provides a robust solution for sea fog removal.
• The multiple scattering model is crucial for handling dense fog scenarios, especially in ocean environments.
• The integration of advanced techniques and multiple loss functions enhances overall performance and image fidelity.