BM3D算法在海洋SAR图像去噪中的应用  

Application of BM3D Algorithm in Ocean SAR Image Denoising

作　　者：石健[1] 汪洋[1] 黄海风[1] 余安喜[1] 李威[2] 

机构地区：[1]国防科技大学电子科学与工程学院,湖南长沙410073 [2]卫星海洋环境动力学国家重点实验室,浙江杭州310012

出　　处：《雷达科学与技术》2016年第1期24-32,共9页Radar Science and Technology

基　　金：上海航天科技创新基金(No.SAST201324)

摘　　要：相干斑噪声一直是困扰SAR图像处理的重要问题,噪声往往将有用信号完全掩盖。由于海洋图像邻域相似性高,将三维块匹配(BM3D)算法应用于海洋SAR图像相干斑噪声的滤除中,能够得到较好的去噪效果。该算法是基于变换域方法和非局部思想发展而来的,通过相似图像块匹配分组生成三维矩阵,然后在三维变换域去噪,最后逆变换还原图像。通过对BM3D算法的深入研究,发现该算法去噪性能与其参数密切相关。针对海洋SAR图像,通过大量实验,总结BM3D参数选择原则,优化了算法的去噪性能。Speckle noise has always been the difficult problem of SAR image processing, the useful information is often covered by the noise. Due to the high similarity of ocean image, BM3D algorithm is applied to the ocean SAR image denoising, and preferable effect can be obtained. This algorithm is based on the transform domain method and non-local approach. It generates a three-dimensional （3D） matrix through block-matching, then denoise at the 3D transform domain, and finally gets the image through inverse transform. By means of studying BM3D algorithm deeply, this paper discovers that its performance is closely related to the parameters. Based on the ocean SAR image, this paper makes a large number of experiments, arrives at a conclusion of the parameter choice, and optimizes the denoising performance of BM3D algorithm.

关 键 词：合成孔径雷达 相干斑噪声 三维块匹配 参数选择 

分 类 号：TN957[电子电信—信号与信息处理] TN958[电子电信—信息与通信工程]