How can I identify number of cells in this image?

I would like to count the number of cells in this image.

[Mohammad Sami]

See if these examples / answers are able to help you.

https://www.mathworks.com/help/images/detect-and-measure-circular-objects-in-an-image.html

https://www.mathworks.com/help/images/identifying-round-objects.html

https://www.mathworks.com/matlabcentral/answers/505196-filling-in-cell-area-and-counting-cells

[AAS]

I have tried all of these.. they fail when there are clusters. Also the black dots within the cells seem to affect the counting too.

Answer 1

[Star Strider]

Maybe I’m missing something in your objection to using those example, however this actually appears to work, even with the clusters. It’s not perfect (I douibt if it ever could be), however it appears to me to be acceptably close:

I = imread('AAS image.png');
[T,EM] = graythresh(I);
BW = imbinarize(I,0.4);
[Ctrs,Radi] = imfindcircles(BW(:,:,1), [9 20],'ObjectPolarity','dark', 'Sensitivity',0.925);

figure
imshow(I)
hold on
h = viscircles(Ctrs,Radi);
hold off
title(sprintf('Showing %d Detected Cells',numel(Radi)))

producing:

.

[AAS]

hi @Star Strider, that is really close. Thank you very much! I was thinking the black dots is whats messing up the image counting. Is there any way to color them or mask them such that they dont get counted?

[Star Strider]

As always, my pleasure!

I can’t think of any way to get rid of the dark dots, other than possibly thresholding them. (I didn’t explore that.) To get an idea of what their magnitudes are, this could help:

figure
surf(I(:,:,1), 'EdgeColor','none')
grid on
colormap(jet(255))
colorbar

Rotate that in the figure GUI to examine it more closely. I’m not certain that it would be straightforward to eliminate them, however.

[Mohammad Sami]

One way can be to use the built in matlab apps like color thresholder to figure out the thresholds you need.

Answer 2

[Image Analyst]

I would first just threshold to exclude the background produce a binary image, and then label the blobs. Then identify blobs which are just single blobs - not touching others. Then get the average area of those. Then sum the binary image and divide by the average area. It should be reasonably accurate, expecially considering the count will vary a lot as the field of view moved or cells enter and leave the field of view. See my Image Segmentation Tutorial in my File Exchange for a demo.

Image Analyst's File Exchange

[AAS]

This is a great idea. Thank you!!