Single image mixed dehazing method based on numerical iterative model and DehazeNet.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Haze significantly degrades image quality and impacts computer vision systems.
• Existing single image dehazing methods, including physical model-based and deep learning-based approaches, struggle to achieve both fidelity and effective haze removal simultaneously in real-world scenarios.

Purpose of the Study:

• To propose a novel mixed iterative model for single image dehazing that combines physical and learning-based methods.
• To enhance image quality by accurately calculating atmospheric light and transmission, even in complex hazy conditions.
• To achieve high-fidelity dehazing that maintains natural image attributes and effectively removes haze.

Main Methods:

• A mixed iterative model integrating physical and learning-based techniques is proposed.
• Images are segmented into regions based on haze density to accurately estimate atmospheric light.
• Joint estimation of transmission using dark channel prior and DehazeNet, followed by numerical iteration to optimize parameters.

Main Results:

• The proposed method demonstrates superior performance compared to state-of-the-art methods on both synthetic and real-world datasets.
• The model effectively removes haze while preserving natural image attributes.
• Visually satisfactory results were achieved when applied to remote sensing datasets, indicating method universality.

Conclusions:

• The mixed iterative model offers a robust solution for high-fidelity single image dehazing.
• The proposed approach successfully addresses the limitations of existing methods in real-world hazy scenes.
• The method's effectiveness extends to diverse applications, including remote sensing image enhancement.