基于图像区域分割的SAR图像去噪算法  

Speckle Filtering of SAR Images Based on Image Segmentation

作　　者：于俊朋[1,2] 尚士泽[1,2] 

机构地区：[1]中国电子科技集团公司智能感知技术重点实验室,南京210039 [2]南京电子技术研究所,南京210039

出　　处：《现代雷达》2016年第9期37-40,共4页Modern Radar

基　　金：中国博士后科学基金面上资助项目(2016M591938)

摘　　要：给出了一种结合图像分割的合成孔径雷达(SAR)图像去噪算法,利用水平集图像分割方法将SAR图像分割得到多个连通区域,并利用基于结构相似性指数的非局部均值滤波(NLM-SSIM)去噪算法对每个连通区域进行去噪。对每个连通域分别去噪利于维持连通区域边缘的原有数值特征,同时也能够保证图像平滑区域的滤波效果,提高了去噪算法的性能。实验部分使用了合成孔径雷达图像中的道路、农田、沟壑和建筑图像块进行测试,将本文算法与非局部均值滤波(NLM)和NLM-SSIM算法进行了去噪效果比较,并通过等效视数(ENL)和边缘平均梯度比(EGR)评价指标验证了文中算法的有效性。A method of speckle filtering of SAR images based on image segmentation is presented. The method of level set is applied to segment SAR images to obtain connected domains. The denoising method of NLM-SSIM is applied to filter speckles in each connected domain. Filtering speckles in each connected domain is propitious to maintain the edge characteristics in image. Simultaneously, it has high effectiveness on smooth region filtering. In experiments, the SAR images of road, farmland, gully, architecture are used to test. The proposed method is compared with NLM and NLM-SSIM filtering to test the denoising effectiveness. Finally, the evaluation criterion of speckle filtering which consist of ENL and EGR are used to demonstrate the performance of proposed method.

关 键 词：合成孔径雷达图像去噪 水平集图像分割 NLM-SSIM去噪 

分 类 号：TN957.52[电子电信—信号与信息处理]