| 引用本文: | 刘超,张晓晖,胡清平.超低照度下微光图像增强神经网络损失函数设计分析.[J].国防科技大学学报,2018,40(4):67-73.[点击复制] |
| LIU Chao,ZHANG Xiaohui,HU Qingping.Design and analysis of loss functions of low-light level image enhancement neural networks under extreme low-light illumination[J].Journal of National University of Defense Technology,2018,40(4):67-73[点击复制] |
|
| |
|
|
| 本文已被：浏览 7057次 下载 7126次 |
| 超低照度下微光图像增强神经网络损失函数设计分析 |
| 刘超1,2, 张晓晖1, 胡清平1 |
|
(1. 海军工程大学 兵器工程学院， 湖北 武汉 430033;2. 军事科学院 系统工程研究院， 北京 100044)
|
| 摘要: |
| 超低照度下（环境照度小于2×10-3lux）微光图像具有低信噪比、低对比度等特点，使目标难以辨识，严重影响观察效果。为了提高超低照度下微光图像质量，设计了一种用于微光图像增强的卷积自编码深度神经网络，并针对传统的均方误差损失函数不符合人类视觉感知特性等问题，结合现有的全参考图像质量评价指标，研究了包括感知损失在内的几种损失函数，并提出了一种新的可微分损失函数。实验结果表明，在网络结构不发生改变的情况下，所提损失函数具有更好的性能，在提高微光图像信噪比和对比度的同时，能够有效地增强图像内部细节信息。 |
| 关键词: 微光图像 图像增强 卷积神经网络 损失函数 |
| DOI：10.11887/j.cn.201804011 |
| 投稿日期：2017-06-20 |
| 基金项目:国家部委基金资助项目（427210843） |
|
| Design and analysis of loss functions of low-light level image enhancement neural networks under extreme low-light illumination |
| LIU Chao1,2, ZHANG Xiaohui1, HU Qingping1 |
|
(1. College of Weaponry Engineering, Naval University of Engineering, Wuhan 430033, China;2.
2. System Engineering Research Institute, Academy of Military Science, Beijing 100044, China)
|
| Abstract: |
| Under the extreme LLL （low light level） conditions (environment illumination less than 2×10-3lux), the LLL image has the characteristics of low signal to noise ratio and low contrast, so that the target is difficult to be identified, thus seriously affecting the observation effect. In order to improve the LLL image quality, a convolutional autoencoder deep neural network for image enhancement was designed. In view of the fact that the traditional mean square error loss function cannot meet the human visual perception characteristics, several loss functions including perceptual loss were studied and a novel, differentiable loss function was proposed in combination with the existing full reference image quality evaluation index. Experimental results show that the proposed loss function can improve the detail information of the image while improving the signal to noise ratio and contrast ratio of the low light level image when the network structure does not change. |
| Keywords: low-light level image image enhancement convolutional neural network loss function |
|
|
