基于Shearlet域系数处理的SAR图像降噪  

SAR image denoising via the process of shearlet coefficients

作　　者：刘书君[1] 吴国庆[1] 张新征[1] 徐礼培 

机构地区：[1]重庆大学通信工程学院,重庆400044

出　　处：《系统工程与电子技术》2015年第9期2023-2028,共6页Systems Engineering and Electronics

基　　金：国家自然科学基金(61301224);重庆市自然科学基金(cstc2012jjA40001);中央高校基本科研业务费(CDJZR11160003;CDJZR12160014)资助课题

摘　　要：结合图像在Shearlet域中系数的特点,提出了一种基于Shearlet系数稀疏表示与投影总变分(total variation,TV)相结合的合成孔径雷达(synthetic aperture radar,SAR)图像去噪算法。有效解决了稀疏表示在图像去噪时存在的边缘细节损失与TV去噪时存在的光滑区域阶梯效应。首先,利用SAR图像Shearlet系数的稀疏性,结合系数稀疏表示模型,采用分段正交匹配追踪方法求解优化解,从统计意义上实现稀疏表示后的系数均值为真实图像系数均值的无偏估计;其次,为弥补稀疏表示中丢失部分系数在图像细节上的损失,同时结合这部分系数对应的Shearlet函数有利于表征图像边缘细节的特性,针对图像在丢失系数对应的Shearlet函数空间中投影重构的结果,结合TV方法迭代去噪。实验结果表明,该方法充分利用Shearlet域系数的特性,采用稀疏去噪与投影TV相结合的方法以弥补各自缺陷,在去噪的同时能有效保持图像纹理细节,并具有更优的图像视觉效果。Combined with the characteristics of the coefficients in the Shearlet domain, an synthetic aperture radar （SAR） image denoising method is presented based on the sparse representation of coefficients and projected total variation （TV） method. The problem that the edge details of the image often lost in the processing of sparse representation and the staircasing effects caused by total variation can be resolved by the proposed method. Firstly, the sparse representation model of the SAR image is constructed and the stagewise orthogonal matching pursuit （STOMP） is used to obtain the optimization solution, which is the unbiased estimation of the real image＇s coefficients in terms of the statistical mean. Secondly, to make up the loss of the image details resuit from the coefficients dropped in the sparse representation processing, the projected total variation scheme is given to iterative denoising, that utilizes the property of the dropped coefficients have the ability to characterize edges of Shearlet coefficients and projects the image to the Shearlet functions that corresponding to these dropped coefficients to get the reconstructed image. The experimental results demonstrate that the proposed method combines the sparse denoising and projection TV based on the characteristics of coefficients in Shearlet domain, that corrects their respective defaults not only suppresses the speckle but also achieves better performance in terms of effectively maintaining the image texture details and subjective visual quality.

关 键 词：合成孔径雷达 图像去噪 稀疏表示 总变分 

分 类 号：TN957.52[电子电信—信号与信息处理]