Low-Light Image Enhancement Techniques: A Review

Abstract

Low-light images from low-quality environments suffer from noise, contrast, and visibility problems, and occur in both computer vision and human perception. Conventional image enhancement methods, including Retinex-based algorithms and histogram equalization, are insufficient for suppressing noise and preserving enhancement in low-light conditions. In recent years, many deep learning-based methods (both hybrids, GANs, and CNNs) have been proposed and have exhibited promising results in enhancing the quality of low-light images. And finally, in the frame of this work, a detailed description of well-known low-light image enhancement algorithms is presented. Their advantages, disadvantages, and applications are discussed. The paper also examines how well these techniques generalize to other domains such as autonomy, surveillance, and medical imaging, and outlines several future directions for research on low-light enhancement.