Structure-transferring edge-enhanced grid dehazing network.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Haze significantly degrades image quality by reducing sharpness due to atmospheric scattering.
• Existing image dehazing methods often suffer from issues like poor brightness, color distortion, artifacts, and blurring.

Purpose of the Study:

• To propose a novel dehazing method that addresses the limitations of previous approaches.
• To enhance image quality by effectively removing haze while preserving structural information and edge details.

Main Methods:

• A Structure-transferring Edge-enhanced Grid Dehazing Network (SEGDNet) is proposed.
• Images are decomposed into low-frequency structure and high-frequency edges using a guided filter.
• Specialized subnetworks handle low-frequency dehazing (LGDSn), high-frequency edge enhancement (HEESn), and fusion (L&HFSn).

Main Results:

• The proposed SEGDNet effectively removes haze, preserves low-frequency structures, and enhances high-frequency details.
• Experimental results on synthetic and real-world datasets show superior performance compared to state-of-the-art methods.
• The method achieves better qualitative and quantitative evaluations in image dehazing.

Conclusions:

• SEGDNet offers a robust solution for image dehazing, overcoming common shortcomings of existing techniques.
• The network's architecture effectively balances structure preservation and edge enhancement for improved visual quality.
• The proposed method demonstrates significant advancements in computer vision for haze removal applications.