I left it to stew for a couple of days and think I found a possible solution for step 2.

I could create two separate images using the second otsu class- the green one in my second image, and the highest class, the one with the maxima.
Then I could use the technique you suggested in your post on corresponding labeled objects in two images to find only those cells that have two corresponding highs from the second image.

Makes sense?

I’d still like your comments on my strategy as a whole if you had a chance. Thank you very much. Arni

I was inspired by this great post and by your cell segmentation post to try something.

I have a number of images that have features very similar to those cells. They are not cells but the analogy is helpful so i will call them so.

All “cells” have a nucleus, but some cells are isolated, some are partially joined.

I have an example here.

What I would like to do is write code to automatically relabel and separate cells with 2 nuclei from ones with only one nucleus. Then only on the former perform watershed and also use bwboundaries to trace the edge as you do in this post.

I developed a strategy but I am stack with one of the steps.

here’s what i’d do:
1 Perform multiclass Otsu segmentation to label image
2 Relabel differently cells with 2 centers from ones with one centre
3 Convert nucleus of isolated cells to hole and use flood fill operation so they become background
4 Perform watershed (or maybe use distance function) on complement of joined cells (hopefully the only ones left)so that nuclei are minima and cell body is maxima.
5) use bwboundaries to trace the edge

I have below an example result of the Otsu multiclass segmentation from step 1 and it looks promising.

Step 3-5 in theory are not difficult to implement.

But I am really stuck with step 2. Do you have any recommendation/suggestion as to the approach I might use?

Or can you think of a simpler solution altogether.

Thank you very much.

Arni

Otsu segmented image

could you tell me how can construct with the image overlay, which is widely
used, e.g. in fMRI. Please have a look at these examples,

http://www.sph.sc.edu/comd/rorden/overlay.html

The color areas are the activated areas determined from other images
acquired during activation and these areas are overlaid to gray
images (background).

thank you very much.