3D Segmentation - Statistical Region Merging 3D

Hi all,

I’m about to segment 3D image stacks by global thresholds to binary images in Fiji/ImageJ 1.51n. For preprocessing, I tried a couple of 3D filters and now I stumbled upon the promising tool ‘Statistical Region Merging’ (SRM) based on the algorithm by Nock et al., 2004. See the ImageJ article.. This results in quite reasonable segmentation results in multiple gray levels (these are the regions), which can later be merged .

To come to my questions:
Applying SRM on a stack, a window pops up with two checkboxes: ‘showAvergages’ and ‘3d’. What about this ‘3d’ option in here? Is there any documentation available? Will SRM not only merge regions within a single slice but also between multiple slices (so z-direction)?

Background: I have confocal microscopy image stacks with irregular cell structures overflowing multiple slices. I just want to extract the whole 3D cell structures from the backround pixels (+reduce some noise) but I don’t want to segment the slices independently. Instead it would be helpfull to consider also the neighbouring pixels in z-direction when classifying a pixel as background or cell structure. That’s why this ‘3d’ option in SRM caught my attention. (In fact, I’m also considering a 3D Weka as segmentation method and using SRM to generate binaries for training…to apply it on bigger data.)

Any reference/experience on 3D SRM? Any help would be appreciated.

Many Thanks

I didn’t find much documentation online, but you can have a look at the comments in the source code:

Looking at the source code, it appears to do true 3D region merging. But why don’t you simply try it on a 3D stack with and without the option checked, and have a look at the differences?


Hello @cbe and welcome to the ImageJ forum!

To add up to Jan’s suggestions, it would be great if you could complete the plugin wiki page with whichever useful information you find out about the plugin’s 3D capabilities :wink:


Cool, thanks for the source code tip @imagejan. I’ll give an update in the wiki on what I’ve found out :wink:

1 Like