Superpixel Segmentation of Polarimetric Synthetic Aperture Radar (SAR) Images Based on Generalized Mean Shift
Superpixel Segmentation of Polarimetric Synthetic Aperture Radar (SAR) Images Based on Generalized Mean Shift
复制标题
基于广义均值平移的极化合成孔径雷达（SAR）图像超像素分割
DOI：
10.3390/rs10101592
复制
发表时间：
2018-10-01
期刊：
影响因子：
5
通讯作者：
Qin, Fachao
中科院分区：
文献类型：
--
作者：
Lang, Fengkai;Yang, Jie;Qin, Fachao
均值移位算法在光学图像分割中表现良好。然而，传统的均值移位算法由于动态范围大，散斑噪声强，直接应用于合成孔径雷达（SAR）图像时，效果较差。近年来，针对极化SAR （PolSAR）图像滤波问题，提出了一种自适应变不对称带宽的广义平均移位（GMS）算法。本文进一步发展了GMS算法用于PolSAR图像分割。基于GMS算法，导出了在联合空间距离域中定义的新的合并谓词。在GMS分割算法中引入了预排序策略和后处理步骤。该算法可直接用于PolSAR图像的超像素分割，无需任何预处理步骤。在机载SAR （AirSAR）和实验SAR (ESAR) l波段PolSAR数据上进行的实验验证了所提出的超像素分割算法的有效性。最后对GMS算法的参数设置、稳定性、质量和效率进行了讨论。
The mean shift algorithm has been shown to perform well in optical image segmentation. However, the conventional mean shift algorithm performs poorly if it is directly used with Synthetic Aperture Radar (SAR) images due to the large dynamic range and strong speckle noise. Recently, the Generalized Mean Shift (GMS) algorithm with an adaptive variable asymmetric bandwidth has been proposed for Polarimetric SAR (PolSAR) image filtering. In this paper, the GMS algorithm is further developed for PolSAR image segmentation. A new merging predicate that is defined in the joint spatial-range domain is derived based on the GMS algorithm. A pre-sorting strategy and a post-processing step are also introduced into the GMS segmentation algorithm. The proposed algorithm can be directly used for PolSAR image superpixel segmentation without any pre-processing steps. Experiments using Airborne SAR (AirSAR) and Experimental SAR (ESAR) L-band PolSAR data demonstrate the effectiveness of the proposed superpixel segmentation algorithm. The parameter settings, stability, quality, and efficiency of the GMS algorithm are also discussed at the end of this paper.