<p>In the field of image enhancement, low-light image enhancement is a notoriously challenging problem. Generally, images captured under poor lightening conditions exhibit characteristics such as color distortions, narrow gray ranges, considerable noise, low brightness, and low contrast. The presence of these characteristics may adversely affect the quality of the visual experience for the human eye and severely limit the performance of machine vision systems. Low-light image enhancement is intended to enhance the visual appearance of such images in order to facilitate subsequent processing. Over the years, many state-of-the-art traditional and deep learning based methods have been developed. This study presents a comprehensive review of these methods, focusing in particular on examining the generalizability of well-known deep learning based approaches to a variety of datasets and evaluation metrics. Based on publicly available datasets, we evaluate the existing methods qualitatively and quantitatively and describe which evaluation metrics perform better. The findings of this experimental review can serve as a reference for future studies, as well as promote the advancement of this area of research. Furthermore, this review discusses the emerging applications, identifies the challenges that remain unsolved, and suggests direction for future research in the field of low-light image enhancement.</p>