基于非局部平均的SAR图像相干斑抑制算法研究

发布时间：2018-05-01 04:17

本文选题：合成孔径雷达图像 + 相干斑　；参考：《西安工程大学》2016年硕士论文

【摘要】：合成孔径雷达(Synthetic Aperture Radar,SAR)是一种相干成像雷达系统,其相干成像的产物SAR图像中随机分布着具有乘性噪声特性的相干斑。相干斑的存在大大降低了解译系统对SAR图像的处理效果,因此SAR图像相干斑抑制成为SAR图像处理研究领域的重要内容之一。目前众多学者在空域、变换域、偏微分各向异性扩散方面提出了多种抑斑算法,然而大部分算法都采用的是局部平均技术。为此,本文从非局部平均(Non Local Means,NLM)技术入手,对SAR图像相干斑抑制方法展开研究。本文围绕非局部平均的SAR图像相干斑抑制开展了以下五个方面的研究:其一,通过分析传统NLM算法的特点,提出了经典NLM算法的快速实现方法。该方法以内存空间为代价,在保持算法去噪效果一致的同时有效地提高了算法的实时性。其二,通过将传统NLM思想引入到经典空域SAR图像抑斑算法中,提出了两种基于空域相似性度量的SAR图像非局部平均抑斑算法。算法一为基于变差系数的SAR图像非局部平均抑斑算法。该算法利用传统NLM技术估计当前像素点与其邻域内其它像素点为中心的相似窗的变差系数比与均值比,用两者的乘积替代权值中高斯加权欧几里得距离,再结合Frost滤波的负指数形式及衰减因子对经典NLM算法中权值计算方法进行了改进,最后利用加权平均实现SAR图像相干斑抑制。算法二为基于Kuan滤波系数的SAR图像非局部平均抑斑算法。该算法通过Kuan滤波对SAR图像预处理,计算信息熵得到Kuan滤波系数,并由传统非局部平均技术估计当前像素点与其邻域内其它像素点为中心的相似窗的均值比,用两者的乘积替代权值中高斯加权欧几里得距离,再结合Frost滤波的负指数形式及衰减因子共同建立新权值,最后通过加权平均实现SAR图像相干斑抑制。两种算法在同质区抑斑效果较好,且算法二较算法一在同质区的平滑性能得到增强,边缘区域细节信息保护能力有所提高。其三,通过将传统非局部平均思想与小波域SAR图像抑斑算法结合,提出了基于小波域相似性度量的SAR图像非局部平均抑斑算法。该算法首先利用小波分解提取SAR图像的低频信息,计算其信息熵。然后利用非局部平均方法估计当前像素点与其邻域内其它像素点为中心的相似窗的均值比,两者结合替代权值中高斯加权欧几里得距离。并与Frost滤波的负指数形式及衰减因子共同建立新权值,最后利用加权平均使SAR图像相干斑得到有效抑制。通过实验图像与数据的分析说明该算法在同质区有较好的平滑能力,且在边缘细节保持方面有较大的突破。其四,借助变尺度窗与基于小波域相似性度量的SAR图像非局部平均抑斑算法,提出了基于变尺度窗的SAR图像非局部平均抑斑算法。该算法通过基于高斯伽马窗的ESM对含噪图像进行阈值化完成图像的区域划分,结合文中搜索窗与相似窗的大小对滤波效果影响的分析,对同质区、边缘区分别采用不同尺度窗,最后通过基于小波域相似性度量的SAR图像非局部平均抑斑算法来抑制相干斑。该算法在同质区得到较好的平滑的同时,保护了边缘细节信息。通过真实SAR图像去噪的实验结果也充分说明,其边缘保持性能较其他算法得到了增强。其五,借助Lab VIEW平台,通过MATLAB与Lab VIEW混编,实现了SAR图像相干斑抑制算法GUI软件。该软件利用本文几种算法对SAR图像进行抑斑处理,不但可以得到相应的去噪图像和边缘检测图像,而且可以得到抑斑后的性能评估参数。本文首先介绍了SAR图像的研究背景及成像原理,阐述了传统非局部平均算法,提出其快速实现方法,然后在空域滤波方面提出了两种算法,在小波域提出一种算法与一种改进算法,最后通过Lab VIEW实现SAR图像抑斑算法GUI软件,并对全文进行了总结与展望。
[Abstract]:Synthetic Aperture Radar (SAR) is a coherent imaging radar system. The coherent imaging products are randomly distributed in the SAR image with multiplicative noise characteristics. The presence of the speckle greatly reduces the processing effect of the interpretation system on the SAR image. Therefore, the SAR image speckle suppression becomes SAR image processing research. One of the important contents of the field is one of the important fields. At present, many scholars have proposed a variety of speckle suppression algorithms in the field of space, transform domain and partial differential anisotropic diffusion. However, most of the algorithms are using local mean technology. Therefore, this paper starts with the non local mean (Non Local Means, NLM) technique to study the speckle suppression method of SAR image. In this paper, the research on speckle suppression of non local average SAR image is studied in the following five aspects: first, by analyzing the characteristics of the traditional NLM algorithm, a fast implementation method of the classical NLM algorithm is proposed. By introducing the traditional NLM idea into the classical spatial domain SAR image suppression algorithm, two non local average speckle suppression algorithms for SAR images based on the spatial similarity measure are proposed. The first is a non local average suppression algorithm for the SAR image based on the coefficient of variation. The algorithm uses the traditional NLM technology to estimate the current pixels and other images in their neighbourhood. The difference coefficient ratio and the mean ratio of the pixel center similar window are replaced by the product of the weight value of the Gauss weighted Euclidean distance in the weight value. Then the weight calculation method in the classical NLM algorithm is improved by the negative exponential form and attenuation factor of Frost filtering. Finally, the SAR image speckle suppression is realized by the weighted average. The second is the second algorithm. Based on the Kuan filter coefficient, the non local average anti speckle algorithm of SAR image is used to preprocess the SAR image by Kuan filtering, and the information entropy is calculated to obtain the Kuan filter coefficient, and the mean ratio of the current pixel points to the other pixels in the neighborhood is estimated by the traditional non local mean technique, and the product is used to replace the weight value in the weight value. Gauss weighted Euclidean distance, and then combined with the negative exponential form and attenuation factor of Frost filter to establish new weights, and finally realized SAR image speckle suppression by weighted average. The two algorithms have better spot suppression in homogeneity area, and algorithm two is better than algorithm one in homogeneity area, and the detail information protection of edge area is protected. Thirdly, by combining the traditional non local mean idea with the SAR image suppression algorithm in the wavelet domain, a non local mean suppression algorithm of SAR image based on the similarity measurement of the wavelet domain is proposed. The algorithm first uses the wavelet decomposition to extract the low frequency information of the SAR image, and calculates its information entropy. Then, the non local mean method is used. The mean ratio of the current pixel to the other pixels in the neighborhood is estimated, and both of them combine the Gauss weighted Euclidean distance in the weight value of the Gauss weighted Euclidean distance. And the new weights are established with the negative exponential form and attenuation factor of the Frost filter. Finally, the weighted average is used to effectively suppress the speckle of the SAR image. The analysis of the data shows that the algorithm has better smoothness in the homogeneity area and has a great breakthrough in the maintenance of edge details. Fourthly, the non local average suppression algorithm of SAR image based on variable scale window is proposed by means of variable scale window and the similarity measurement based on wavelet domain. The algorithm is based on the variable scale window. Based on Gauss gamma window ESM, the image is divided by thresholding of the noisy image, and the influence of the search window and the size of the similar window on the filtering effect is analyzed. The different scale windows are used in the homogeneous area and the edge region. Finally, the non local average suppression algorithm of the SAR image based on the similarity of the wavelet domain is used to suppress the image. The algorithm protects the edge details while getting better smoothness in the homogeneity region. The experimental results of denoising through real SAR image are also fully explained, and its edge retention performance is enhanced compared with other algorithms. 5, with the help of Lab VIEW platform, MATLAB and Lab VIEW are mixed, the SAR image speckle suppression algorithm GU is realized. I software. This software uses several algorithms to suppress the SAR image, not only can get the corresponding denoising image and edge detection image, but also get the performance evaluation parameters after the speckle reduction. This paper first introduces the research background and imaging principle of the SAR image, and expounds the traditional non local average algorithm, and proposes its fast reality. In this way, two algorithms are proposed in the field of spatial domain filtering, and an algorithm and an improved algorithm are proposed in the wavelet domain. Finally, the SAR image suppression algorithm GUI software is implemented by Lab VIEW, and the full text is summarized and prospected.

【学位授予单位】：西安工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN957.52

【相似文献】