融合边缘信息的合成孔径雷达图像超像素分割算法

针对简单线性迭代聚类(simple linear iterative cluste, SLIC)对含有乘性相干斑噪声的合成孔径雷达(synthetic aperture radar, SAR)图像边缘分割不理想的问题，本文在SLIC基础上提出了一种融合边缘信息的SAR图像超像素分割算法。首先，利用高斯方向平滑对SAR图像进行预处理，从而在抑制乘性相干斑噪声的同时有效保护边缘细节；其次，提出了一种基于指数加权平均比率(ratio of exponential weighted average, ROEWA)算子的改进相似度测量参量，以提高SAR图像的分割精度；最后，采用六边形初始化聚类中心与圆形区域的搜索方式进行局部区域聚类，从而保证了算法复杂度增加的同时，算法的运行时间不会明显变化。实验结果表明，与四种经典超像素算法相比，本文算法生成的超像素边缘更加贴合SAR图像的真实边缘且得到的超像素大小较为均匀。

SAR image superpixel segmentation algorithm based on edge information fusion

Aiming at the problem that simple linear iterative cluste superpixel algorithm is not ideal for edge segmentation of synthetic aperture radar images with multiplicative coherent speckle noise, a SAR image superpixel segmentation algorithm fusing edge information is proposed based on SLIC in this paper. Firstly, the SAR image is pre-processed with Gaussian direction smoothing to effectively protect edge details while suppressing multiplicative coherent speckle noise. Secondly, an improved similarity measurement parameter is proposed based on the ROEWA operator, which improves the accuracy of the SAR image segmentation. Finally, local area clustering is performed by a hexagonal initialization cluster center and circular area search method, thereby ensuring that the algorithm complexity does not increase while the algorithm running time does not change significantly. The experimental results show that compared with the four classic superpixel methods, the superpixel edges generated by the algorithm in this paper, which is more closely fit the true edges of SAR images and the obtained superpixel sizes are more uniform.