HistRetinex: Optimizing Retinex Model in Histogram Domain for Efficient Low-Light Image Enhancement

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Retinex-based low-light image enhancement methods offer excellent performance but are often time-consuming for large images.
• Existing methods face challenges with computational efficiency, limiting their application in real-time scenarios.

Purpose of the Study:

• To develop a novel, fast low-light image enhancement method by extending the Retinex model to the histogram domain.
• To address the computational inefficiency of traditional Retinex models for large-scale image processing.

Main Methods:

• Proposed HistRetinex, a histogram-based Retinex model that approximates Retinex decomposition in the histogram domain.
• Developed a two-level optimization model based on prior information and the histogram-based Retinex model.
• Derived iterative formulas for illumination and reflectance histograms and enhanced images by histogram matching.

Main Results:

• HistRetinex demonstrated acceptable estimation errors between predicted and actual low-light image histograms.
• Outperformed existing unsupervised enhancement methods in visibility and performance metrics.
• Achieved a processing time of 1.85 seconds for a $1000\times 664$ image, significantly faster than traditional methods.

Conclusions:

• HistRetinex offers a computationally efficient and effective solution for low-light image enhancement.
• The histogram-domain approach successfully accelerates Retinex-based image enhancement.
• Accelerated implementation using MATLAB MEX achieved 28 FPS for $600\times 400$ images, enabling real-time applications.