Segmentation of roughly circular objects in an image.

Hello everybody,

I've been trying to segment this image for some time, and chiefly I need to identify the nuclei as signified by dark, roughly circular objects. Original image (Photobucket)

My current train of thought uses a Canny edge detect, and then a circular Hough transform from there: Edge detected image. However, as you can see in my edge detection, there are a lot of edges which clearly don't correspond to nuclei, and I'm wondering if it would improve things significantly if I could discard the edges which are too long or too short to be a nucleus. To that end, is it possible to find the length of each 'chain' which is an edge, and discard them?

That's my main question, but if anybody has any suggestions that might help the problem as a whole, then that would be fantastic too!

Many thanks, Callum

Answer 1

[Image Analyst]

Normally we don't give advice on algorithms, since this is a MATLAB forum, but you might try a sigma filter or a morphological erosion to get rid of the white clutter. Then try to flatten the image with CLAHE using adapthisteq(). Then try thresholding, labeling, and calling regionprops. Throw out blobs that aren't round by checking their circularity (perimeter^2/(4*pi*area)). Or you could try imfindcircle and see how that does. Try Brett's GUI to make it easy: http://www.mathworks.com/matlabcentral/fileexchange/34365-findcirclesgui

Answer 2

[Callum]

Thank you very much for your answer - I'll try those things and see where I get; they all sound promising. Apologies - I did not realise that questions about algorithms aren't really appropriate here. For future reference, is there a better place that I should ask?

[Image Analyst]

Well simple ones are okay but you look like you have a really tough problem that's not going to be solved in a few lines of code or a few minutes of my time. So you can go to the sci.image.processing newsgroup, or try VisionBib to search published literature on image processing: http://iris.usc.edu/Vision-Notes/bibliography/contents.html