低光图像增强方法综述  

A REVIEW OF LOW-LIGHT IMAGE ENHANCEMENT METHODS

作　　者：孙建德 孙晓燕[1] 张瑞瑞[2] 李静[1] 高玲[1] Sun Jiande;Sun Xiaoyan;Zhang Ruirui;Li Jing;Gao Ling(School of lnformation Science and Engineering,Shandong Normal University,250358,Jinan,China;Research Center of Intelligent Equipment,Beijing Academy of Agriculture and Forestry Science,100000,Beijing,China)

机构地区：[1]山东师范大学信息科学与工程学院,济南250358 [2]北京市农林科学院智能装备技术研究中心,北京100000

出　　处：《山东师范大学学报(自然科学版)》2025年第1期1-20,共20页Journal of Shandong Normal University(Natural Science Edition)

基　　金：山东省自然科学基金联合基金重点资助项目(ZR2022LZH012)。

摘　　要：在低光照条件下拍摄的图像会严重影响视觉辨识能力。低光图像增强就是解决低光照图像系列退化问题的方法,可以有效地提高人眼和机器视觉对于低光图像的辨识和理解能力。传统的低光照图像增强方法通常需要根据不同场景设计特定的先验知识,而这些先验知识的推导过程往往具有局限性,不适于广泛的实际应用。随着大规模数据集的诞生,基于深度学习的低光图像增强方法已成为计算机视觉领域备受关注的研究课题之一。本文对低光图像增强领域的研究及进展进行全面综述,通过系统地总结和分类增强方法与数据集,分析当前研究的主要挑战与技术难点,并进一步探讨了该领域未来的发展方向与潜在的研究趋势,旨在为后续研究提供有价值的参考与启示。Images captured under low-light conditions severly degrade visual.Addressing a series of degradation issues in low-light images can effectively improve both human and machine vision system's ability to interpret and understand low-light images.Traditional low-light image enhancement methods typically require designing specific prior knowledge for different scenarios,and the derivation process of this prior knowledge is often highly complex,making it impractical for real-world applications.With the advent of large-scale datasets,deep learning-based low-light image enhancement methods have become one of the most popular and significant research topics in the field of computer vision.This paper aims to provide a comprehensive review of the latest research advancements in low-light imageenhancement,systematically summarizing and categorizing the related enhancement methods and datasets.It also analyzes the main challenges and technical difficulties in current research.Furthermore,this paper explores future development directions and potential research trends in this field,aiming to offer valuable references and insights for subsequent research.

关 键 词：低光图像增强 图像重构 低级视觉任务 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]