采用GSM模型进行稀疏表示的SAR图像降斑算法

发布时间：2018-04-25 22:27

  本文选题：高斯比例混合模型 + 同步稀疏编码　； 参考：《信号处理》2017年11期

【摘要】：针对SAR图像降斑过程中会产生过平滑现象及相干斑的滤除不彻底等问题,提出了稀疏结构符合高斯比例混合(Gaussian Scale Mixture,GSM)模型的SAR图像降斑算法。根据贝叶斯原理以及相干斑的统计特性推导该算法的数学模型,在块匹配过程中使用概率而不是欧式距离进行权重衡量,根据图像块之间的结构相似度,可以有效区分同质区与异质区,并得到图像块的较优均值估计。使用PCA字典学习方法对每个图像块进行子字典训练,实现同步稀疏编码(Simultaneous Sparse Coding,SSC),数学模型的求解利用迭代正则化方法。分别使用合成场景SAR图像及真实场景SAR图像对算法进行验证,实验表明,相比于目前已提出的PPB算法、SAR-BM3D算法及FANS算法,该算法能有效提高等效视数,在滤除相干斑的同时很好地保留了图像的局部结构特性与纹理特征。
[Abstract]:In order to overcome the problems of over-smoothing and incomplete speckle filtering in SAR image, a new algorithm for speckle reduction in SAR image is proposed, which accords with Gao Si's proportional mixing Gaussian Scale mixture (GSM) model. According to the Bayesian principle and the statistical characteristics of speckle, the mathematical model of the algorithm is deduced. In the process of block matching, the weight is measured by probability instead of Euclidean distance. The homogeneous region and the heterogeneous region can be effectively distinguished, and the optimal mean estimation of the image block can be obtained. The PCA dictionary learning method is used to train the sub-dictionary of each image block to realize the synchronous sparse coding Simultaneous Sparse coding and the iterative regularization method is used to solve the mathematical model. The algorithm is verified by synthetic scene SAR image and real scene SAR image respectively. The experimental results show that the algorithm can effectively improve the equivalent visual number compared with the existing PPB algorithm and FANS algorithm. The local structure and texture features of the image are well preserved while the speckle is removed.
【作者单位】： 南京航空航天大学自动化学院;
【基金】：国家自然科学基金项目(61501228)
【分类号】：TN957.52
，

本文编号：1803271