Color and Luminance Separated Enhancement for Low-Light Images with Brightness Guidance.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Existing Retinex-based methods for low-light image enhancement often use complex networks, leading to imprecise estimations.
• Limitations in current strategies hinder effective restoration and enhancement of low-light images.

Purpose of the Study:

• To introduce a straightforward and effective strategy for Retinex decomposition.
• To improve low-light image enhancement by refining reflectance and illumination map estimations.

Main Methods:

• Developed a novel Retinex decomposition approach by dividing images into colormaps and graymaps.
• Employed a diffusion model for separate enhancement of reflectance and illumination maps.
• Integrated brightness guidance into illumination map enhancement for perturbation removal and brightness adjustment.

Main Results:

• The proposed method achieved a performance improvement of approximately 4.4% on the LOL dataset compared to state-of-the-art diffusion-based methods.
• Demonstrated superior performance in quantitative and qualitative experimental analyses.
• Validated the model's generalizability across multiple real-world datasets.

Conclusions:

• The novel Retinex decomposition strategy offers a more effective approach to low-light image enhancement.
• The method provides improved restoration and brightness control, outperforming existing techniques.
• The approach shows promise for practical applications in various real-world scenarios.