用于MR图像非局域均值去噪的边缘增强策略

非局域均值(NLM)滤波有很好的去噪效果并已成功地应用于磁共振图像的去噪中，但与所有去噪方法相同，总是会在一定程度上模糊图像细节. 该文提出将从原始图像中提取出来的高频信息与NLM去噪图像相融合，来还原在去噪过程中丢失的细节. 首先利用一种基于拉普拉斯金字塔的多分辨率方法，从原始图像中提取出包含丰富的边缘信息的高频组分. 然后利用作者提出的一种新的基于SUSAN算子的边缘检测算子产生一幅连续的边缘图，并利用该边缘图将高频组分与NLM方法去噪的图像相融合. 该方法在图像的平滑区域取得了良好的去噪效果，同时可以保留甚至增强图像的细节. 同时，该方法对图像的增强不会导致增强图像中常见的伪影.

An Edge Enhancing Scheme for Non-Local Means Denoised MR Images

Non-Local Means (NLM) is an excellent filter for noise removal and has been applied successfully to MR images. However, just like every other denoising filters, it inevitably blurs fine structures in images to certain extent. In this paper, we propose to combine high frequency components (HFC) extracted from the original image with the NLM-denoised image to recover the lost details. High frequency components, which contain rich edge information, are extracted with a multi resolution method based on Laplacian Pyramid. Then the HFCs are combined with the image denoised by NLM using a continuous edge map produced by an edge detector which is based on SUSAN operator and modified to be more robust to noises. The experiment showed that the proposed scheme achieved good suppression of noise in smooth regions while preserves or even enhances image details. Furthermore, images were enhanced subtly without artifacts that are typical in enhanced images.