一种自适应的非局部均值图像去噪算法

AN ADAPTIVE NON LOCAL MEANS IMAGE DENOISING ALGORITHM

对Buades等人提出的非局部均值图像去噪算法进行改进。传统的方法在滤波参数定义上存在缺陷,为了解决这个问题,通过建立噪声方差与滤波系数的关系,提出解决噪声估计的方法。另外,根据小波系数的分布特点,利用GGD模型参数(尺度和形状参数)对系数进行拟合,并用GGD模型参数提出一种有效的噪声方差估计算法。实验结果表明,该噪声方差估计算法不仅能有效地估计噪声方差大小,而且使原有的非局部均值算法具有自适应性。这种自适应的非局部均值算法可以达到近似最优,具有鲁棒性和快速性,且算法精度高。

In this paper, we make the improvements on non-local means （NL-Means） algorithm introduced by Buades et al. Original NL- Means algorithm has the defect in filtering parameter definition. In order to solve this problem, we present the solution for noise estimation by establishing the relation between noise variance and the filtering parameter. Besides, according to the distribution feature of wavelet coeffi- cients, the Coefficients are fitted by using the generalised Gaussian distribution （GGD） model parameters （scale and shape parameters）. We also propose an effective noise variance estimation method using GGD model parameters. Experimental results show that the noise variance es- timation method can effectively estimate the size of noise variance, it can also makes the original NL-means algorithm adaptive. Such adaptive NL-Means algorithm can reach approximately optimal value, and has robustness and fastness with high accuracy.