基于非局部相似模型的多尺度几何变换域SAR图像去噪

发布时间：2018-03-04 21:03

本文选题：SAR图像去噪　切入点：剪切波变换　出处：《河北大学》2017年硕士论文　论文类型：学位论文

【摘要】：合成孔径雷达(Synthetic aperture radar,SAR)是通过采用合成孔径的理论,得到高分辩率的微波成像,起初主要应用在机载和星载平台,随着科学技术的不断完善,出现了弹载、无人机SAR、以及手持式设备等多种形式平台搭载的合成孔径雷达,广泛用于军事、民用领域。由于SAR图像是通过接收地物散射回波的相干信号叠加而形成的,因此极易受到相干噪声的干扰,从而严重的影响了SAR图像的后续处理,例如目标提取、目标分类等。所以,SAR图像相干噪声的抑制方法具有十分重大的研究价值。由于多尺度几何变换不仅具有良好的时频特性,而且能够更好的表示图像信息,因此其在SAR图像去噪中得到了普遍的使用。另外,由于图像自身存在着冗余性,从而导致图像的像素块之间存在着相似性,该性质称之为图像的非局部相似性。很多学者将图像的非局部相似性应用到SAR图像去噪中,取得了十分好的结果。本人根据前人的研究,在此基础上从下面两个方面进行深入的探究。一方面,将已有的多尺度几何变换的性质进行提高并创新出的拥有更多性质的多尺度几何变换。另一方面,本文在分析了多尺度几何变换优势的基础上,结合图像的非局部相似模型对SAR图像的相干噪声进行抑制。论文的主要研究工作如下:1.基于剪切复小波变换的SAR图像去噪算法研究根据剪切波变换缺少方向选择性和移不变性的缺陷,文章结合剪切波和双树复小波的各自特点,提出了剪切复小波变换。该变换与非下采样复小波变换一样保留了多个低频子带,但是具有更好的稀疏性,因此可以更好的表示图像信息。然后,本文利用该变换结合K-SVD去噪算法、软阈值去噪和Lee滤波提出了一种新的SAR图像去噪方法。与其它去噪方法相比,此方法不仅考虑了SAR图像高频部分的去噪,还对SAR图像的低频部分进行了噪声抑制,仿真效果显示了所提出的去噪算法的真实性和可靠性。2.基于权重优化的广义非局部均值SAR图像去噪为了克服变换域算法产生的人造纹理和非局部均值算法权重选择的难题,从而更好的抑制相干噪声和尽可能多的保留图像的纹理信息,文章结合非下采样剪切波变换和广义非局部均值算法提出了一种基于非下采样剪切波域权重优化的广义非局部均值SAR图像去噪方法。在权重优化的过程中,本文算法利用重叠子块中的冗余信息,采用二次规划求解最小均方误差意义下的最优权重。通过实验对比,本文的方法具有较好地去噪结果。
[Abstract]:Synthetic aperture radar (SAR) is to obtain high resolution microwave imaging by using the theory of synthetic aperture. At first, it was mainly used in airborne and spaceborne platforms. With the development of science and technology, projectile was produced. Synthetic Aperture Radar (SAR) carried by UAV, hand-held equipment and other platforms is widely used in military and civil fields. Because SAR images are formed by the superposition of coherent signals received from scattered echoes from ground objects. Therefore, it is easy to be interfered by coherent noise, which seriously affects the subsequent processing of SAR image, such as target extraction. Therefore, the suppression method of coherent noise in SAR images has great research value. Multi-scale geometric transform not only has good time-frequency characteristics, but also can better represent the image information. Therefore, it is widely used in SAR image denoising. In addition, because of the redundancy of the image itself, there is similarity between the pixel blocks of the image. This property is called non-local similarity of image. Many scholars apply non-local similarity of image to SAR image denoising, and obtain very good results. On the one hand, the properties of the existing multi-scale geometric transformation are improved and the multi-scale geometric transformation with more properties is innovated. On the other hand, In this paper, the advantages of multi-scale geometric transformation are analyzed. Combining the nonlocal similarity model of the image, the coherent noise of the SAR image is suppressed. The main research work of this paper is as follows: 1. The SAR image denoising algorithm based on the shearing complex wavelet transform is studied according to the lack of direction selection of the shear wave transform. The imperfections of sex and shift invariance, Combined with the characteristics of shearing wave and bitree complex wavelet, a shearing complex wavelet transform is proposed in this paper, which preserves many low frequency subbands as well as non-downsampling complex wavelet transform, but has better sparsity. Therefore, the image information can be represented better. Then, this paper proposes a new SAR image denoising method by using the transform combined with K-SVD denoising algorithm, soft threshold denoising and Lee filtering. Compared with other denoising methods, this paper proposes a new SAR image denoising method. This method not only takes into account the high frequency part of SAR image de-noising, but also carries on the noise suppression to the low frequency part of the SAR image. Simulation results show the authenticity and reliability of the proposed denoising algorithm. Secondly, the generalized non-local mean SAR image denoising based on weight optimization is used to overcome the problem of artificial texture and weight selection of the non-local mean algorithm, which is generated by the transform domain algorithm. In order to better suppress coherent noise and retain as much texture information as possible, In this paper, a generalized nonlocal mean SAR image denoising method based on non-downsampling shear wave domain weight optimization is proposed by combining non-downsampling shear wave transform and generalized non-local mean algorithm. In this paper, the optimal weight in the sense of minimum mean square error is solved by quadratic programming using redundant information in overlapped subblocks. The experimental results show that the method presented in this paper has better denoising results.
【学位授予单位】：河北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52

【相似文献】