光学遥感图像变量分裂迭代快速复原算法

Variable Splitting Iterative Fast Algorithm for Remote Sensing Image Recovery

为了实现光学遥感图像的快速复原,分析了已有全变差正则化复原算法存在的问题,提出基于全变差正则化图像复原的代理代价函数模型,并将该模型转化为3个子问题优化过程,设计了一种快速解析迭代的变量分裂算法。同时为了克服"阶梯效应",考虑人眼对图像平坦区域和边缘区域噪声的感知特性,给出了迭代系统中正则化参数的自适应估计。实验结果表明:本文算法在信噪比改善和计算时间都优于Wiener滤波、约束最小二乘、基于全变差梯度下降和交错子空间投影等算法,保留了图像的细节信息,减少了"寄生波纹效应"和"阶梯效应"。

In order to realize the fast deblurring for optical remote sensing images,the disadvantages of the original total variation regularization（TV） based algorithm are studied.A surrogate cost functional based on TV model is proposed.The proposed model was first transformed into three minimizing sub-problems with close form solution and then a variable splitting based fast algorithm is presented.Meanwhile,to overcome the ＂staircase effect＂,the regularization parameters are estimated and adaptively adjusted by perceptual sensitivity to different areas such as flat and edge areas.Experiments show that the proposed algorithm is superior to state-of-the-art algorithms such as Wiener filter,restricted least-squares filter,TV based gradient decreasing algorithm and alternative subspace projection algorithm with respect to signal to noise ratio and the computing time.In addition,both the Gibbs effect and staircase effect are reduced by the proposed algorithm without the loss of image information.