非局部主成分分析极大似然估计MRI图像Rician噪声去噪

Maximum Likelihood Estimation Image Denoising Using Non-Local Principle Component Analysis

由于MRI图像中噪声呈Rician分布,直接使用现有针对高斯噪声的去噪方法将引入误差。基于此本研究使用Rician噪声模型改进现有极大似然估计去噪的高斯模型,同时引入非局部主成分分析,在非局部区域选择灰度和纹理均具有较高相似性的像素进行最优复原估计。使用非局部主成分分析不仅克服现有局部性去噪方法模糊边界的缺陷,而且具有更高的图像细节信息复原能力。分别应用所提出的方法、局部极大似然估计去除Rician噪声方法、采用参数修正非局部均值去除Rician噪声方法、无特定噪声模型的全变差方法,对不同噪声等级和不同纹理复杂度的图像进行定性和定量的去噪实验。结果表明,所提出的方法可在保持图像细节和纹理信息的前提下有效去噪,较之现有方法效果更好。

As the noise in the MRI are under the Rician distribution, implementation of commonly used denoising methods which is designed for Gaussian noise will introduce bias. This paper used the Rician noise model instead of the Gaussian noise model which is currently used in the maximum likelihood estimation, and used the non-local principle component analysis to obtain optimal estimation by choosing pixels with high similarity for both texture and gray level. The proposed method can improve the drawback of boundary blurring with the non local principle component analysis and increase the ability of restoring detail information with the maximum likelihood estimation. Experiments compared with the proposed method, local maximum likelihood estimation, parameter correction non-local mean and total variation method were implemented in different noise standard and different geometric complex MRI images quantitatively and qualitatively. The result demonstrated that the proposed method can remove the noise effectively together with the plausible detail information and texture compared with the currently used methods.