TUSR-Net: Triple Unfolding Single Image Dehazing With Self-Regularization and Dual Feature to Pixel Attention

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Single image dehazing is challenging due to information loss in hazy conditions.
• Deep learning methods often neglect the low similarity between clear and haze components.
• Existing approaches are limited by the lack of contrastive constraints between image components.

Purpose of the Study:

• To propose an end-to-end self-regularized network (TUSR-Net) for improved single image dehazing.
• To leverage the contrastive peculiarity of image components for better dehazing performance.
• To enhance feature representation and fusion for more effective image restoration.

Main Methods:

• Developed TUSR-Net, an end-to-end network utilizing self-regularization (SR).
• Separated hazy images into clear and hazy components, applying SR to guide recovery.
• Employed a triple unfolding framework with dual feature-to-pixel attention for intermediate information fusion.
• Incorporated a weight-sharing strategy for parameter efficiency and flexibility.

Main Results:

• TUSR-Net effectively exploits contrastive peculiarities between image components.
• The proposed method significantly improves performance in single image dehazing.
• Achieved a superior trade-off between performance and parameter size.
• Demonstrated state-of-the-art results on various benchmarking datasets.

Conclusions:

• TUSR-Net offers a novel and effective approach to single image dehazing.
• The self-regularization strategy and attention mechanisms enhance image restoration quality.
• The network provides a flexible and efficient solution for practical applications.