A Survey of Low-light Image Enhancement Algorithms based on Deep Learning

Low-light image enhancement (LLIE) represents a critical challenge in computer vision, focusing on improving image visibility while preserving fine details and natural color fidelity. While traditional approaches like histogram equalization and Retinex-based methods have been widely used, they often suffer from noise amplification and color distortion artifacts. Recent advances in deep learning have revolutionized this field, achieving unprecedented performance. However, the rapid proliferation and diverse nature of these deep learning techniques necessitate a comprehensive and systematic review to consolidate current knowledge, structure the evolving landscape, and identify key trends and challenges. This paper systematically reviews state-of-the-art deep learning methods for LLIE, including convolutional neural network (CNN)-based architectures, generative adversarial networks (GANs), and Transformer models. Furthermore, we comprehensively analyze benchmark datasets, standardized evaluation metrics, and persistent challenges in the field, while also outlining promising directions for future research.