基于曲率带的超分辨率图像复原参数估计

Curvature Bounds-based Parameter Estimation in Super-Resolution Image Restoration

在很多超分辨率复原应用中,正则化参数是未知的。然而通过L-曲线估计正则化参数的计算代价十分昂贵。本文提出在超分辨率复原中使用基于Lanczos算法和Gauss积分的方法来高效计算正则化参数。该方法用Gauss积分来计算矩阵矩,通过部分Lancros算法来计算L曲线的曲率带。该方法可减少正则化参数的计算代价和确定Lancros算法的恰当迭代次数。

In many super-resolution image restoration applications, the regular parameters are not known. However, the computationa for the estimation of the unknown parameters is quite costly though L-curve. The paper proposes an efficient method based on the Lanczos algorithm and Gauss quadrature rules in super- resolution image restorations. The method can compute matrix moments though Gauss quadrature rules. L-curvature ribbon is yielded with partial Lanczos algorithm. The method can reduce the computational complexity of the regular parameters and determine the suitable iterative times for the Lanczos algorithm.