夜晚图像增强方法综述  

Summary of nighttime image enhancement methods

作　　者：姜雪松 姚鸿勋[1] JIANG Xuesong;YAO Hongxun(School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《智能计算机与应用》2020年第3期394-403,共10页Intelligent Computer and Applications

基　　金：国家自然科学基金(61472103,61772158,61702136)。

摘　　要：夜晚图像增强作为图像增强领域的一个重要分支,一直受到学者们的关注。夜晚图像增强的目标是提升夜晚图像的整体质量,包括提升亮度和细节可见度、抑制噪声等方面。本文收集了大量真实夜晚图像,并通过人工合成大量低照度图像,使用这些数据对多种夜晚图像增强方法进行分析,比较各种方法在夜晚图像增强任务中的表现,对这些方法的原理、优点和存在的不足加以分析。研究使用主观评价和客观评价两种图像质量评价方法同时评价不同方法的性能,主观评价从视觉感知方向直接评价增强图像的质量,客观评价方法则分为有参考的图像质量评价和无参考的图像质量评价,本文分别使用生成低照度图像和真实夜晚图像测试所有算法性能,为了保证对比实验的公正性,所有算法的参数都按照原论文参数设置。分析和实验结果表明,基于图像处理方法的夜晚图像增强方法能够有效提升图像亮度,但也发现了传统图像处理在夜晚图像增强问题中存在的瓶颈,而基于机器学习的夜晚图像增强方法能够改变夜晚图像的亮度分布情况,更好地提升夜晚图像质量,夜晚图像增强的进一步发展必然使用机器学习方法。Nighttime image enhancement,as an important branch of image enhancement,has attracted the attention of researchers.The goal of nighttime image enhancement is to improve the overall quality of night image,including enhancing brightness and detail visibility,suppressing noise and so on.Firstly,this paper collects a large number of real night images and synthesizes a large number of low-light images artificially.Then,the paper uses these data to analyze various nighttime image enhancement methods,compares their performance,and analyzes the principles,advantages and disadvantages of these methods.After that,the paper uses subjective evaluation and objective evaluation to evaluate the performance of different methods at the same time.Subjective evaluation directly evaluates the quality of enhanced images from the direction of visual perception.The objective evaluation methods are divided into reference image quality evaluation and non-reference image quality evaluation.Low-light is used respectively.Degree image and real night image test the performance of all algorithms.In order to ensure the fairness of the contrast experiment,all the parameters of the algorithms are applied to set according to the parameters of the original paper.The analysis and experimental results show that the night image enhancement method based on image processing method can effectively improve the brightness of the image,but it also finds the bottleneck of the traditional image processing in the night image enhancement.The night image enhancement method based on machine learning can change the brightness distribution of the night image and improve the night image quality better.The further development of night image enhancement will inevitably use machine learning method.

关 键 词：夜晚图像增强 图像处理 Retinex模型 图像分解 深度学习 

分 类 号：TU375[建筑科学—结构工程]