基于几何特征聚类的SAR图像降噪

合成孔径雷达（SAR）图像中的相干斑噪声严重干扰了图像的自动解译。本文结合非局部滤波技术，提出了一种新的基于图像块处理的滤波方法。该算法首先提取SAR图像的几何特征，然后根据几何特征对图像块进行聚类，根据乘性噪声模型，对具有同一类几何特征的图像块进行非局部加权滤波处理，在充分利用图像块的强度相似性的同时，保留了SAR图像的几何结构特征。仿真图像和真实SAR图像的实验结果证实了本文算法的有效性。

Speckle Reduction in SAR Images Based on Geometric Clustering

Speckles, also known as multiplicative noises in synthetic aperture radar(SAR) images make it difficult to achieve auto-interpretation of the images. This paper proposes a novel patch-based noise reduction strategy based on the non-local method. Firstly, the algorithm extracts the geometrical features of the image, and then applies improved non-local weighted denoising method to the patches with the same structural features. It takes advantage of photometrical similarity among the patches as well as reserves geometrical features of the images. Experiments in both simulated and real SAR images demonstrate the high performance of the proposed algorithm.