基于Curvelet域隐马尔可夫树模型的SAR图像去噪  

SAR Image De-noising Based on Curvelet Domain Hidden Markov Tree Models

作　　者：金海燕[1] 焦李成[1] 刘芳[2] 

机构地区：[1]西安电子科技大学智能信息处理研究所雷达信号处理国家重点实验室,西安710071 [2]西安电子科技大学计算机学院,西安710071

出　　处：《计算机学报》2007年第3期491-497,共7页Chinese Journal of Computers

基　　金：国家"九七三"重点基础研究发展规划项目基金(2001CB309403);国防预研基金(A1420060172);国家自然科学基金(60472084)资助

摘　　要：从SAR图像相干斑噪声的统计特点出发，将Curvelet变换与隐马尔可夫树（HMT）模型相结合，提出了一种基于Curvelet域隐马尔可夫树（HMT）模型的图像去噪方法．利用HMT模型捕获Curvelet系数之间的尺度从属性，较好地实现了普通图像去噪和SAR图像的相干斑噪声抑制，同时分析了文中算法的去噪机理和计算复杂度．仿真实验证明，与小波域HMT模型方法和Curvelet变换方法比较，主观视觉效果和数值指标都有明显改进．平滑指数（FI）值大小适中，水平和垂直边缘保持指数（ESI）平均提高了约0．2～0.3．Based on the statistical property of SAR image speckle noise and combining curvelet transform with HMT models, a method of SAR image de-noising based on curvelet domain hidden Markov tree （HMT） models is presented in this paper. Using HMT models to capture the scale dependencies among curvelet coefficients, it implements the image de-noising and reduces SAR Speckle noise effectively, furthermore, analyze the algorithm mechanism and computation complexity. De-noising performance is evaluated through subjective inspection, as well as objective measurements-flatness index and edge save index. Results clearly demonstrate the superiority of this new approach when compared to conventional wavelet domain HMT and curvelet transform. FI value obtained is appropriate and ESI vertically and horizontally are averagely increased by about from 0.2 to 0. 3.

关 键 词：CURVELET变换 HMT模型 RIDGELET变换 SAR图像 图像去噪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]