一种新的遥感影像组合滤波去噪方法  

A new combined filtering and denoising algorithm for remote sensing image

作　　者：任超 张胜国[1,3] 李现广 王子彦 闫志恒 REN Chao;ZHANG Shengguo;LI Xianguang;WANG Ziyan;YAN Zhiheng(College of Geomatics and Geoinformation,Guilin University of Technology,Guilin 541006,China;Guangxi Key Laboratory of Spatial Information and Geomatics,Guilin University of Technology,Guilin 541006,China;PowerChina Guiyang Engineering Corporation Litimed,Guiyang 550081,China)

机构地区：[1]桂林理工大学测绘地理信息学院,广西桂林541006 [2]桂林理工大学广西空间信息与测绘重点实验室,广西桂林541006 [3]中国电建集团贵阳勘测设计研究院有限公司,贵阳550081

出　　处：《桂林理工大学学报》2023年第3期442-449,共8页Journal of Guilin University of Technology

基　　金：国家自然科学基金项目(42064003)。

摘　　要：针对目前对影像中高斯噪声与椒盐噪声的去噪算法普遍存在去噪效果不佳、去噪后影像边缘模糊的问题,提出了一种新的遥感影像组合滤波去噪方法。该方法在DnCNN算法的基础上增加了一层自适应中值滤波,利用自适应中值滤波对椒盐噪声去噪效果好的特点去除影像中残留的噪声,为了在去噪过程中更好地保护影像细节、边缘等信息,在自适应中值滤波过程中采用噪点及其最近邻域像素点中值替换原滤波窗口中值,提升细节信息保护能力,同时对影像进行边界扩展,确保原影像边界像素点自适应参与滤波处理,提高影像边缘去噪效果,使去噪后影像更加清晰。为验证算法的可行性,分别进行了仿真和真实遥感影像去噪实验,并用主观视觉和客观评价指标进行对比分析。结果表明,本文方法对于遥感影像混合噪声去噪效果优于传统去噪方法,能够较好地保留图像边缘及纹理细节。Remote sensing image denoising is of great significance for the subsequent application and research of images.Gaussian noise,salt and pepper noise are common noises in images.Current denoising algorithms generally have disadvantages such as poor denoising effect for this type of mixed noise and blurry edges of the image after denoising.To solove these problems,a new combined filtering and denoising method for remote sensing images is proposed which adds a layer of adaptive median filter on the basis of DnCNN algorithm.Adaptive median filter is to remove the residual noise in the image with good denoising effect on salt and pepper noise.In order to get better image details,edges and other information,the median value of noise points and their nearest neighbor pixels are used to replace the median value of the original filter window to improve the protection of detail information in the adaptive median filtering process.At the same time,the boundary of the image is expanded to ensure the boundary of the original image pixels adaptively participate in filtering processing to improve the denoising effect of the image edge and make the image clearer after denoising.In order to verify the feasibility of the algorithm,a simulation experiment and a real remote sensing image denoising experiment were carried out,and comparative analysis was done based on subjective vision and objective evaluation indicators.Experimental results show that the algorithm proposed in this paper is better than traditional denoising methods for mixed noise denoising in remote sensing images.It can betterly preserve the image edges and texture details.

关 键 词：遥感影像 去噪 DnCNN算法 自适应中值滤波 边界扩展 

分 类 号：P407.8[天文地球—大气科学及气象学]