论文总字数：19554字

目 录

1 引言 1

2 BM3D算法及其SAR图像去噪算法 3

2.1 BM3D 3

2.1.1 BM3D算法思想..........................................................................................................3

2.1.2 BM3D算法原理..........................................................................................................3

2.2 适应SAR图像去斑的BM3D 5

3 SAR去噪改进算法 6

3.1 块相似性度量 6

3.2 群组收缩 7

3.3 聚合 10

4 快速自适应SAR去噪声（FANS）算法 10

4.1 FANS算法介绍 11

4.2 FANS算法的改动 11

5 实验结果 13

5.1 图像质量评价方法 13

5.2基于BM3D的SAR图像去噪算法结果 13

6 结论和展望 20

参考文献 21

致谢 22

基于BM3D理论的SAR图像去斑点噪声方法

顾宇

，China

Abstract: With the rapid development of times, the information technology is constantly improving, and the image is getting a lot of attentions because of its convenience and information. Due to the influence of external and internal factors, the collected images not only contain the information that people want, but also contain some noises. Therefore, image denoising technology becomes one main research topic. In many cases, image denoising is the key to the processing of the problem, such as high-speed imaging and radar imaging. Many denoising techniques are developed, and most of them aim at the additive white gaussian noise model, in addition, some are designed for the multiplicative noise model. However, there are few denoising techniques for different types of noise. This paper deals with the above-stated issue.

Combining wavelet transform theory with non-local thought, a despeckle methodfor SAR image is proposed. This method is based onBM3D theory for additive gaussian white noise. Due to the particularity of SAR image, some steps of the original BM3D algorithm are modified. The process of block matching probability similarity measure is applied, and wavelet transformfor additive noise signal related model is used. Thus an optimal local linear minimum mean square error estimate is obtained, and the improvements of SAR image denoisingcan be better.Compared with several advanced methods, the proposed method has better results in both signal-to-noise ratio and perceived image quality.

Key words:Wiener filtering;synthetic aperture radar (SAR); despeckle; nonlocal filtering;

1 引言

由于相干散射现象的存在，合成孔径雷达（SAR）图像固定的受到斑点噪声的影响。尽管斑点本身带有关于照明区域的信息，但它降低了图像的外观并影响了由计算机程序或由人类解释器执行的场景分析任务的性能。为了解决这些问题，人们经常采用多视角技术，即非连续的计算独立图像中某些数的平均值，从而降低了噪声强度，但这种方法同时丢失了图像信息。因此，发掘更合适的滤波技术显得尤为重要，这种技术既能够在减少噪声影响的同时，也能够更好的保留图像的特征信息，如辐射测量和纹理信息等边缘细节。

早期的一些去斑点噪声技术使用所说的同态方法，这种方法的主要思想是取数据的对数，来获得一个比较容易处理的图像模型，然后运用在加性高斯白噪声（AWGN）去噪中提出的那些人们比较熟知的方法。虽然这些方法具有简单性这一不可否认的优点，但它忽略了散斑的一些基本性质。因为实际上，通过对数变换后的斑点是非高斯的，有着非零均值，一般来说，在执行任何操作之前都需要校正偏差。更严重的是，对数变换从根源上改变了数据的动态，导致在去噪过程中无法避免的发生放射性畸变，使图像增加了其他一些影响图像质量的因素。

剩余内容已隐藏，请支付后下载全文，论文总字数：19554字