The Non-Local Means Filter is a very powerful filter to reduce noise, while still keeping edges sharp. To learn more about the underlying theory click here.

The size of edges is controlled with the Patch Radius that determines the size of regions that are compared. The Search Window Radius determines the neighborhood around each voxel, for which regions are compared. The amount of smoothing is controlled by the Filter Strength. For very noisy images it can be necessary to Enable Prefilter. By applying a Gaussian filter, noise is reduced before applying the Non-Local Means Filter. This option then improves the patch comparison.

The left image shows the original gray values of an image. The right image shows the same image after applying the Non-Local Means Filter using default settings.

The Non-Local Means Filter can significantly improve the quality of segmentations. However, applying it to a complete dataset may take longer than the application of other filters. Therefore, we recommend to always use the Preview function before to identify the ideal settings.