基于变量分离和加权最小二乘法的图像复原*

为提高图像复原的质量和速度,提出一种新的图像复原算法。首先基于变量分离技术,加入新的约束条件,建立解决图像复原问题的目标函数;然后利用交替最小化方法,将目标函数的优化分解为两个交替迭代的过程,以获得图像复原问题的全局最优解。在求解分离得到的新变量的过程中,引入迭代重加权最小二乘法(IRLS)处理L1范式的不可微分问题。实验结果表明,提出的算法有效地解决了图像复原问题;与同类的一些算法相比,该算法在复原速度和复原效果方面均具有优势。

Image restoration based on variable splitting and weighted least squares

For improving and accelerating image restoration,this paper proposed a novel algorithm.Based on the variable splitting technology,it established an objective function with a new constraint for image restoration problem.To obtain the glo-bal optimal solution of image restoration,it used the alternating minimization method to decompose the optimization of the objective function into two alternately iterative procedures.In the procedure of calculating the new variable obtained by splitting,it introduced the iteratively reweighted least squares method to solve the non-differentiable L1 norm.The experimental results demonstrate the efficiency of the proposed algorithm,and compared with some state-of-the-art algorithms,it shows better performances on the restored results and speed.