基于经验模态分解和核主成分分析的SAR图像相干斑抑制  

SAR image despeckling based on empirical mode decomposition and kernel principle component analysis

作　　者：李长超[1,2] 王文波[3] 

机构地区：[1]武汉科技大学计算机科学与技术学院,湖北武汉430065 [2]武汉科技大学智能信息处理与实时工业系统湖北省重点实验室,湖北武汉430065 [3]武汉科技大学理学院,湖北武汉430065

出　　处：《武汉科技大学学报》2016年第3期224-230,共7页Journal of Wuhan University of Science and Technology

基　　金：国家自然科学基金资助项目(11201354);卫星海洋环境动力学国家重点实验室资助项目(SOED1405)

摘　　要：结合经验模态分解(EMD)与核主成分分析(KPCA)方法,针对SAR图像提出一种改进的相干斑抑制算法。首先将SAR图像进行对数变换和均值调整后进行经验模态分解;然后利用KPCA进一步去除各层本征模态函数(IMF)中的噪声,具体方法是根据斑点噪声的统计特性和零均值高斯白噪声IMF的能量分布模型,近似计算各层IMF中噪声能量所占比例,据此选择合适数量的主成分重构IMF;最后对经过KPCA处理的IMF进行累加重构得到去噪SAR图像。实验结果表明,与另外两种EMD图像去噪算法相比,本文提出的方法在相干斑抑制效果和图像细节信息保持能力两方面都有较好的提高。This paper proposes a speckle suppression method for SAR image using empirical mode decomposition（EMD）and kernel principle component analysis（KPCA）.Firstly,SAR image after logarithmic transformation and mean adjustment is decomposed by EMD.Then the noises in the intrinsic mode functions（IMF）are removed by KPCA,which is performed as follows：the proportion of noise energy in each IMF is approximately calculated based on the statistical properties of speckle noise and IMF energy distribution model of Gaussian white noise with zero mean,and IMF is reconstructed by selecting the appropriate principle components according to the noise energy proportion.Finally,the denoised SAR image is obtained by accumulating the intrinsic mode functions processed by KPCA.Experimental results show that,compared with two other EMD-based image denoising algorithms,the proposed method has better performance in suppressing speckle noise and preserving detail image information.

关 键 词：SAR图像 相干斑抑制 经验模态分解 核主成分分析 本征模态函数 图像去噪 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]