A survey on learning-based low-light image and video enhancement

Abstract

Low-light enhancement (LLE) is a fundamental technique for improving the visual perception and interpretability of images and videos that suffer from low light degradation. In recent years, learning-based low-light image and video enhancement has made significant strides. Low-light image enhancement (LLIE) mainly focuses on enhancing images in a spatial-varying manner, while low-light video enhancement (LLVE) emphasizes exploiting temporal information in videos. In this survey, we present a comprehensive review of the research progress of LLE, categorizing LLIE and LLVE solutions according to their task attributes for the first time. We also provide a systematic analysis and discussion of technical details from various aspects. To deepen researchers’ understanding of LLE technology development and provide performance benchmarks, we extensively evaluate various LLIE and LLVE models using datasets for low-light image, video, and high-level visual applications. Based on the experimental analysis, we summarize the current limitations and challenges of LLE. Additionally, our study offers insights into potential future research directions for LLE.