基于自适应参数回归的非局部图像滤波算法

针对图像加性高斯白噪声,提出一种优化的自适应参数滤波算法。该算法以非局部欧氏中值(non-local Euclidean medians, NLEM)滤波算法为基础,根据含噪图像梯度幅值在一定噪声范围内服从Rayleigh分布这一特性,求得以梯度幅值和噪声标准差为自变量的二元自适应滤波参数,并将它引入到邻域的权值计算中。其次,噪声的变化影响着l_p范数回归的选择,在一定范围内以噪声标准差为自变量对参数p进行多项式拟合,得到自适应l_p范数回归。在自适应滤波参数基础上,用自适应l_p范数回归进一步改进NLEM滤波算法的l₁范数回归。所选图像的实验结果表明，本文算法在一定噪声范围内不但获得满意的去噪效果,而且有效地减少人机交互程度。

Non-local image filter algorithm based on adaptive parameter regression

For additive white Gaussian noise of an image, this paper proposes an optimized adaptive parameter filter algorithm. Based on the non local Euclidean medians (NLEM) algorithm, according to the property that the noise image gradient amplitude obeys Rayleigh distribution within a certain noise range, we obtain a binary adaptive filter parameter by regarding gradient amplitude and noise standard deviation as independent variables. The adaptive filter parameter is introduced in the weight calculation of neighbors. Furthermore, the changes of the noise affect the selections of the l_p norm regression. Make p used polynomial fit with noise standard deviation in a certain range, and get adaptive l_p norm regression. On the basis of adaptive filter parameters, l₁ norm regression used in NLEM can be improved by using adaptivel_p norm regression. It is verified that the new algorithm not only obtains satisfactory results of denoising in a certain noise range, but also reduces the degree of human-computer interaction effectively.