基于贝叶斯估计的非平稳SAR图像自适应小波滤波

合成孔径雷达(SAR)的相干斑噪声严重影响图像质量、降低图像的可判读性。本文提出的滤波方法，将通用小波阈值滤波方法和基于图像局域统计特性的滤波方法相结合。针对非平稳SAR图像的不同区域采用不同的方法调整小波系数，具有自适应能力。分析了噪声在小波变换域中的分布和统计特性，用高斯混合模型来对其进行描述。采用期望最大似然(EM)算法迭代出小波系数的分布参数。最后用贝叶斯估计得出真实图像的小波系数。在仿真实验中，将该方法与经典的局域统计方法以及通用的小波阈值去噪方法进行了比较。结果表明，本文的方法能对非平稳SAR图像进行更有效的滤波。

A daptive SAR image filtering based on wavelets and Bayesian estimation

Coherent speckle noise of SAR images decreases the image quality seriously,and makes the image difficult to be understood.In this paper,a new filter algorithm using the advantages of both local statistical filter and wavelet threshold filter is introduced.The algorithm was an adaptive one,for the wavelet coefficients were adjusted according to various kinds of regions in a non-stationary SAR image.Being studied carefully,the distribution of noise in the wavelet domain was described by Gauss Mixture Model.The Expectation Method (EM) was used to calculate the distribution parameters of the wavelet coefficients.In the end,the wavelet coefficients of the noiseless SAR image were estimated with the Bayesian estimator.The filter result was compared with that of both the local statistical filters (such as Lee filter) and wavelet threshold filter.Results showed that the method introduced in this paper performed better to various regions,especially to non-stationary SAR images.