改进U-Net神经网络下LII质量增强方法  

Low-Illumination Image Quality Enhancement Method Baed on Improved U-Net Neural Network

作　　者：王玮 董富江[2] WANG Wei;DONG Fu-jiang(Chongqing College of Communication and Information engineering,Chongqing College of Mobile Communication,Chongqing 401520,China;College of Science Ningxia Medical University,Yinchuan Ningxia 750004,China)

机构地区：[1]重庆移通学院通信与信息工程学院,重庆401520 [2]宁夏医科大学理学院,宁夏银川750004

出　　处：《计算机仿真》2024年第9期200-204,共5页Computer Simulation

摘　　要：低照度图像由于光线不足,通常存在过暗、低对比度等问题,质量增强可以提高图像的质量和清晰度,使得图像更容易被计算机识别和处理。但是由于图像邻近像素之间通常存在较高的空间关联性,导致图像增强的难度较高。为此,提出一种改进U-Net神经网络下低照度图像质量增强方法。结合用卷积网络和下采样改进U-Net神经网络,将低照度图像分解为反射和光照两个部分,分别提取两个部分特征,并将其输入到改进的U-Net神经网络内,获取初步重建图像。同时利用Retinex理论对光照部分增强,将初步重建图像和增强处理后的光照分量两者融合,最终实现低照度图像质量增强。经实验测试证明,采用所提方法可以有效改善图像质量,获取满意的图像质量增强效果,且耗时更短。Low-illumination images often suffer from the problems of excessive darkness and low contrast due to insufficient lighting.Quality enhancement can improve the clarity of image,making it easier for computers to recognize and process.However,high spatial correlation between adjacent pixels makes image enhancement more difficult.To address this issue,this article proposed a method for enhancing low-illumination image quality based on improved U-Net neural network.Firstly,we used convolutional networks and down-sampling to improve the U-Net neural network.Then,we divided the low-light image into two parts:reflection and illumination.Secondly,we extracted features from both parts and input them into the improved U-Net neural network to obtain a preliminary reconstructed image.Meanwhile,we enhanced the illumination part by the Retinex theory.Moreover,we integrated the reconstructed image with the enhanced illumination component.Finally,we achieved the enhancement of low-illumination image quality.The experimental results show that the proposed method can effectively improve image quality and obtain satisfactory enhancement effects,with less time.

关 键 词：改进神经网络 低照度图像 特征提取 质量增强 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]