Structure-Revealing Low-Light Image Enhancement Via Robust Retinex Model

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Conventional Retinex models for low-light image enhancement do not account for image noise.
• Noise is an inherent artifact in images captured under low-light conditions, degrading enhancement performance.

Purpose of the Study:

• To propose a robust Retinex model that incorporates a noise map for improved low-light image enhancement.
• To develop an optimization framework addressing noise, illumination, and reflectance for superior image quality.

Main Methods:

• Introduced a robust Retinex model considering illumination, reflectance, and a noise map.
• Formulated an optimization function with novel regularization terms for illumination smoothness and reflectance detail.
• Employed an augmented Lagrange multiplier-based alternating direction minimization algorithm for efficient problem solving.

Main Results:

• The proposed method effectively enhances low-light images, outperforming conventional approaches.
• Demonstrated significant improvements in image quality, preserving structural details while reducing noise.
• Validated the model's effectiveness in challenging conditions like underwater and remote sensing imagery.

Conclusions:

• The robust Retinex model offers a superior solution for low-light image enhancement in the presence of noise.
• The method's adaptability extends to various image enhancement applications, including specialized domains.
• This work advances low-light image processing by integrating noise estimation within the Retinex framework.