Belangrijkste concepten
The author introduces an edge-guided Retinex model for enhancing low-light images using a novel inertial Bregman alternating linearized minimization algorithm.
Samenvatting
The content discusses the challenges of low-light image enhancement, proposes an edge extraction network, analyzes the effectiveness of the proposed approach through experiments, and compares it with state-of-the-art methods. The results show improved performance in enhancing real-world low-light images.
- The proposed method integrates edge information to enhance low-light images effectively.
- Experiments demonstrate the superiority of the proposed scheme over traditional and deep learning-based methods.
- The approach shows robustness in non-reference quality assessment metrics across various datasets.
Statistieken
Generally, a low-light image S can be decomposed as the reflectance part R and illumination part L.
Directly enhancing R and L to reconstruct the desired image ˆS may be challenging as both are unknown.
A total variation model was proposed to preserve edges in image segmentation tasks.
The proposed inertial Bregman alternating linearized minimization algorithm aims to solve structured optimization problems effectively.
Citaten
"The enhanced results with our edge prior have more clear detail."
"Our method achieves the best averaged overall performance in terms of PSNR and SSIM among all methods."