摘要
结合稀疏贝叶斯学习和压缩感知(CS)理论,提出了一种基于贝叶斯匹配追踪的SAR图像重构的新方法。该方法将SAR图像的重构过程看做是一个线性回归问题,而待重建的图像是该回归模型中的未知权值参数。利用高斯混合参数对未知权值参数赋予确定的先验条件概率分布,用于限制权值参数的稀疏性。该方法能够得到重建图像所需要的一组具有较高后验概率密度的模型,从而实现图像在最小均方误差(MMSE)意义下的重构;对于高斯混合模型中参数未知的情况,可以采用基于EM的最大似然估计方法估计。实验结果表明,基于贝叶斯匹配追踪的SAR图像重构方法能够获得精确的重建图像,并且能够有效地保持图像的细节特征。
Based on sparse learning and CS theory,this paper proposed a new SAR image reconstruction method.The process of image reconstruction was treated as a linear regression problem and the image to reconstruction was the unknown parameters of the regression model.It used Gaussian mixture parameters to predefine the prior conditional density of the unknown parameters in order to confine the sparsity.A set of model could be achieved that could be used to reconstruct the image in sense of MMSE.When the hyperparameters were unknown,the method based on EM could be used to estimate.Simulation results indicate that the Bayesian matching pursuit based method can get a precisely reconstructed image and the details can be preserved.
出处
《计算机应用研究》
CSCD
北大核心
2012年第7期2722-2724,2736,共4页
Application Research of Computers
基金
中央高校基础研究基金资助项目(ZYGX2009Z005)
国家自然科学基金资助项目(60772143)
关键词
压缩感知
SAR图像
高斯混合参数
贝叶斯
EM
compressive sensing(CS)
SAR image
Gaussian mixture parameter
Bayesian
EM