基于低通滤波残差图的高光谱条带噪声去除  

Removal of Hyperspectral Stripe Noise Using Low-Pass Filtered Residual Images

作　　者：鞠荟荟 刘志刚 姜江军 汪洋 Ju Huihui;Liu Zhigang;Jiang Jiangjun;Wang Yang(Institute of Nuclear Engineering,Rocket Force Engineering University,Xi'an,Shaanxi 710025,China)

机构地区：[1]火箭军工程大学核工程学院,陕西西安710025

出　　处：《光学学报》2018年第12期353-361,共9页Acta Optica Sinica

基　　金：国家自然科学基金(41574008)

摘　　要：针对高光谱遥感图像中存在的条带噪声,提出了一种基于低通滤波残差图的条带噪声去除算法。算法首先使用高斯低通滤波器对图像进行滤波,得到低通滤波残差图;然后借助条带噪声秩为1以及残差图中的细节与条带噪声正交的先验信息,使用正交子空间投影技术将低通滤波残差图中的条带噪声和图像细节进行分离;最后将分离出的细节信息加入滤波后的图像中。通过对上述三步不断迭代,算法能够有效地去除图像中的条带噪声,并且能够解决低通滤波法去条带造成图像模糊的问题。实验结果表明,与现有前沿的去条带算法相比,该方法能在有效去除条带噪声的同时很好地保持图像的信息。A stripe-removing algorithm using low-pass filtered residual images is proposed herein to remove stripe noise in hyperspectral remote sensing images.First,a Gaussian low-pass filter is used for image filtering to obtain a low-pass filtered residual image.Then,using previously determined knowledge that the rank of the stripe noise is 1and the details are orthogonal to the stripe noise,we employ the orthogonal subspace projection technique to separate the stripe noise from the details in a low-pass filtered residual image.Finally,the separated details are then added to the filtered image.Through continuous iteration of the above mentioned three steps,the proposed algorithm can effectively remove stripe noise and overcome image blurring issues caused by traditional low-pass filtering methods.The experimental results illustrate that the proposed algorithm can significantly improve the removal of stripe noise and preserve image information comparing with the existing stripe-removing algorithms.

关 键 词：遥感 高光谱遥感 条带噪声 低通滤波 残差图 正交子空间投影 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]