Priori Knowledge Makes Low-Light Image Enhancement More Reasonable

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Low-light image enhancement is challenging due to inherent uncertainties in brightness mapping.
• Existing methods struggle with adaptive pixel brightness adjustments and global visual balance.

Purpose of the Study:

• To develop a priori knowledge-based framework for robust low-light image enhancement.
• To address the ill-posed nature of low-light image enhancement by guiding brightness.
• To improve visual balance and detail preservation in enhanced images.

Main Methods:

• Incorporation of priori channels to guide the brightness of enhanced images.
• Development of an enhancement function that adaptively adjusts priori enhancement probability based on pixel brightness.
• Introduction of the Global-Attention Block (GA Block) for inter-pixel computation and visual balance.

Main Results:

• Priori DCE demonstrates significant advantages over state-of-the-art methods on the LOLv2-Synthetic dataset.
• Achieved improvements in PSNR (25.67 to 29.49) and SSIM (92.82 to 93.6) compared to Retinexformer.
• Reduced the NIQE index from 3.94 to 3.91, indicating enhanced perceptual quality.

Conclusions:

• Priori DCE effectively enhances low-light images by leveraging prior knowledge and adaptive strategies.
• The proposed framework offers superior performance in terms of quantitative metrics and perceptual quality.
• The Global-Attention Block is crucial for achieving visual balance in low-light image enhancement.