基于增强的同质相似性和旋转不变性的MRI去噪

MRI denoising using enhanced homogeneous similarity and rotational invariance

提出了一种基于增强的同质相似性和旋转不变性的MRI去噪算法。考虑到图像的旋转不变性和同质相似性,融合了体像素(voxel)的亮度、相应邻域块的均值、梯度以及一致锐度。通过这个融合,块状的结构信息,如边缘、角形、末端能被很好地保护。提出的方法和近年来相关的先进方法在Brainweb数据库进行定量对比,结果表明提出的方法均优于其他方法,在去除噪声的同时能很好地保护图像的细节。最后,将提出的方法应用于OASIS临床数据库。

In this paper, a novel method based on the three- dimensional non- local means(NLM3D)filter is presented to improve the denoising results of three- dimensional magnetic resonance images. Taking rotational invariance and homogeneity similarity into consideration, we build the filter by integrating voxel intensity, and the mean, average of gradient(AG)and region homogeneity(RH)of the corresponding local patch. In this way,the structure information of the patches such as edges, corners and ends can be preserved well. The presented method has been compared with the related state- of- the- art methods over synthetic datasets and the experimental results show that the proposed approach outperforms other methods, and removes noise successfully while preserving details of the images. Finally, we apply our method on the OASIS database.