用于图象复原的离散L-曲线自适应算法

An Adaptive Discrete L-Curve Algorithm for Image Restoration

描述了一种自适应的离散L-曲线正则参数的计算方法,正则化方法为截断奇异值分解法。该算法不需要预置参数,能够滤除L-曲线上的局部特征,略掉局部拐点,确保了解的收敛。在不同的噪声背景下以及不同尺度下我们进行了数值测试,并将测试结果与最优正则解进行对比分析。实验表明,该算法对图像的复原效果、可靠性和稳定性要优于传统的算法。

This paper describes an adaptive Discrete L-curve algorithm to calculate regular parameters. The regularization method based on truncated singular value decomposition without pre-defined parameters. It can filter small local phenomena and local corners to guarantee the convergence of the solution. Numerical examples have been made in different noise level and scales to compare the testing results with the optimal regular solution. The experiments have shown that the proposed algorithm is superior to traditional algorithm ones in terms of image restoration, effect, robustness and stability.