Depth-Guided Bilateral Grid Feature Fusion Network for Dehazing.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Adverse foggy weather degrades image quality, reducing contrast and visibility.
• Traditional dehazing methods struggle in complex real-world environments.
• Existing deep learning approaches often fail to fully utilize depth and edge information, causing artifacts.

Purpose of the Study:

• To propose a novel deep-guided bilateral grid feature fusion dehazing network.
• To address limitations in current single-image dehazing techniques.
• To improve fog removal while preserving image details and edge clarity.

Main Methods:

• A deep-guided bilateral grid feature fusion dehazing network was developed.
• Depth information was extracted and used to guide bilateral grid feature sampling.
• Feature reconstruction and final image estimation involved convolutional layers and residual connections.

Main Results:

• The proposed network effectively removed fog from images in experiments.
• The method demonstrated superior performance on public datasets compared to existing approaches.
• Image details and edge information were successfully preserved post-dehazing.

Conclusions:

• The deep-guided bilateral grid feature fusion network offers an effective solution for single-image dehazing.
• This approach significantly improves image quality in foggy conditions.
• The method shows promise for practical applications requiring clear imagery in adverse weather.