L_1-L_2正则化的图像复原交替优化算法

L_1-L_2regularization based alternating optimal algorithm for image restoration

为快速且准确地重建原始图像,提出一种新的图像复原算法。在稀疏表示的框架下,建立图像复原问题的约束优化模型,同步估计原始图像及其稀疏表示。复原模型的目标函数包含L1-L2双正则项,为此采用交替优化将模型分解为若干子问题,交替迭代求解这些子问题。其中不可微分的子问题,由迭代重加权方法进行处理。实验结果表明,仅需较少次迭代该算法即可获得原始图像及其稀疏表示的最优估计。与某些优秀的同类算法相比,该算法的速度更快,复原图像的质量更高。

For rapidly and accurately restoring images, a novel image restoration algorithm is presented. In the framework of sparse representation, a new constrained optimization model is created, which enables the estimations of the original image and its sparse representation. The objective function of the model has L1-L2 regularized terms, thus the alternating optimization method is introduced to decompose the model into equivalent sub-problems. The non-differentiable one among sub-problems is handled by iterative reweighted method. The experimental results demonstrate that with only a few itera-tions, the presented algorithm can achieve optimal estimations of the original image and its sparse representation. Com-pared with some state-of-the-art algorithms, the presented algorithm shows to be faster, and obtains better results.