Low-light image enhancement with retinex theory optimized by improved whale optimization algorithm

Abstract

Low-light image enhancement remains challenging due to low contrast, blurred details, and color distortion. This work introduces a Retinex-based enhancement framework optimized by an improved Whale Optimization Algorithm (WOA) to automatically tune enhancement parameters. A Cauchy-chaos initialization is adopted to diversify the search and enhance global exploration, while adaptive weight updating and Gaussian perturbations improve local exploitation. By jointly optimizing Retinex parameters–including Gaussian kernel standard deviations (σ1,σ2,σ3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _1, \sigma _2, \sigma _3$$\end{document}), illumination gain, bias, and color factors–within a multi-objective scheme, we enable adaptive parameter tuning for low-light enhancement. The method is evaluated on five public datasets (LOL, MEF, DICM, VV, and NPE) and demonstrates competitive or superior performance compared with conventional enhancement approaches (auto white balance, adaptive contrast enhancement, histogram equalization) and representative deep-learning models (Zero-DCE, SCI, RUAS, EnlightenGAN). Performance gains are quantified via information entropy (IE), average gradient (AG), and perceptual quality indices (NIQE, BRISQUE). An ablation study identifies Gaussian perturbation as the dominant contributor (79.2%), followed by adaptive weight updating (19.6%), with Cauchy initialization contributing the remaining 1.2% under the current configuration. Overall, the proposed WOA-Retinex framework achieves natural-looking, detail-rich enhancements across diverse low-light scenarios.