首页 >  2018, Vol. 22, Issue (4) : 570-580

摘要

全文摘要次数: 1655 全文下载次数: 1312
引用本文:

DOI:

10.11834/jrs.20187418

收稿日期:

2017-10-13

修改日期:

PDF Free   HTML   EndNote   BibTeX
结合相似块匹配及线性最小均方误差滤波器的全极化雷达影像去噪
1.安徽大学 资源与环境工程学院, 合肥 230601;2.高分辨率对地观测系统安徽数据与应用中心, 合肥 230000
摘要:

针对全极化合成孔径雷达PolSAR(Polarimetric Synthetic Aperture Radar)影像相干斑噪声严重的问题,提出了一种结合相似块匹配和线性最小均方误差原理的去噪方法。该方法首先在原始影像上实现相似块组的匹配,进而利用线性最小均方误差滤波器对影像块组进行滤波得到初始去噪结果;然后,同时利用原始影像和初始去噪影像的信息进行相似块组的重新匹配,并再次利用线性最小均方误差原理对重匹配影像块进行去噪,得到影像最终的去噪结果。利用模拟的PolSAR影像和高分三号卫星PolSAR影像进行了算法效果的验证。结果表明,去噪算法在显著抑制影像噪声水平的同时,也能较好地保持影像的边缘和极化特性等细节信息。

Polarimetric radar image despeckling by using similar patches matching and minimum mean square error filter
Abstract:

Synthetic Aperture Radar (SAR) systems are capable of capturing images of the earth in day and night and for almost all weather conditions. Polarimetric SAR (PolSAR), which focuses on emitting and receiving complete polarized radar waves to characterize observed targets, is an advanced form of SAR. PolSAR data have unique advantages in obtaining land cover information in comparison with optical remote sensing data, thereby resulting in their wide use in many areas. However, the SAR data are inherently affected by speckle noise. The presence of speckles complicates the PolSAR image interpretation and land surface parameter inversion. Therefore, despeckling is an essential procedure in most cases before using SAR images to obtain land cover information. In the traditional methods based on the linear minimum mean square error (LMMSE) filter, a group of homogeneous image pixels is first selected in a local window to obtain precise filter parameters in the LMMSE estimator. The LMMSE estimator is then generated from the values of the selected pixels and is saved as the filtered value of the pixel being processed. These methods assume that all of the selected pixels are absolutely homogeneous pixels with respect to the processed pixel, which is insufficiently reliable. In addition, most of these methods have limitations, such as a limited selection range of pixels, which are only compared with the characteristics of their own, thereby possibly producing a biased or inferior estimation of the filter parameters in the estimator.
In this study, we proposed a similar patch matching and LMMSE filter combined with a PolSAR despeckling method based on the 3D block matching-based algorithm. The main idea behind the proposed method is to select additional similar pixels in the nonlocal area to improve the performance of the LMMSE estimator. The main process of the proposed method is as follows:first, the similar patches in a nonlocal window for each target patch are selected to form a patch group, and the LMMSE filter is used to filter all the pixels in the group. Second, an aggregation step is utilized to estimate the pixels that have been clustered into several groups. The patch matching process is used again to group similar patches by considering the information of the original and basic estimated images. Finally, a collaborated LMMSE filter and an aggregation step are undertaken to filter the image.
The experiments on the simulated PolSAR and two real PolSAR images acquired by the GF-3 satellite revealed the positive despeckling performances of the proposed method. The speckle is reduced to a large degree, and the image details, such as the edges and strong point targets, are effectively preserved.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群