l_1-总变分图像恢复的一种快速解法

总变分正则对具有清晰边缘的图像有较高的恢复性能.鉴于其不可微,交替方向法被广泛用来求解该类问题.然而,交替优化子问题常常使得搜索方向不精确.针对交替方向法在l_1-总变分图像恢复模型中的应用,提出了一种利用子空间优化矫正其当前搜索方向的方法.数值实验结果表明推出的方法可以有效地提升图像恢复性能.

A Fast Method For l1 -Total Variation Image Restoration

Abstract: For images with clear edges, total variation regularization has better restoration performance and the alternating direction method is wildly applied to solve these problems due to the nondifferentiability. However, the search directions are not accurate usually because of optimizing subproblems alternatively. One method based on the subspace optimization is presented to corrects the current search direction of the alternating direction method with l1-total variation image restoration model in this paper. Numerical experimental results illustrate that the proposed method could improve the restoration performance effectively.