Low-Light Image Enhancement Using Machine Learning and Retinex-Based Illumination Correction

Abstract

Images captured in poor lighting often appear dark, noisy, and unclear, making it difficult to identify important visual details. This issue affects many real-world applications such as surveillance, autonomous navigation, and photography, where visibility is essential. Traditional techniques like histogram equalisation and gamma correction can brighten images, but often create unwanted noise and unnatural colours. To overcome these limitations, this research introduces a Low Light Enhancing System Using Machine Learning that automatically improves image brightness and colour tone while maintaining natural appearance and detail. The proposed approach uses a deep learning-based model inspired by Retinex theory to separate and enhance illumination and reflectance layers. By training the model on publicly available datasets such as LOL and See-in-the-Dark (SID), the system learns to adapt to various lighting environments effectively. Experimental results show that the enhanced images achieve higher Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) scores compared to conventional methods. The system provides a practical and efficient solution for improving low-light images, contributing to clearer visuals and better performance in vision-related applications. In addition, the proposed system is designed to be fully automated, requiring minimal user intervention while delivering consistent enhancement results across different image types. By leveraging the learning capability of deep neural networks, the model adapts to varying illumination conditions and preserves fine details that are often lost in conventional enhancement methods. This adaptability makes the approach suitable for real-time and practical use cases, where reliable low-light enhancement is critical for accurate visual interpretation and downstream computer vision tasks.