Separating overlapping edges in an image to get an accurate count

Hi,

I am trying to count the big cells in the sample image below.

I was able to reach a point where I was able to delete most of the smaller cells and highlight the big cells as seen below.

However, when I count the objects (using bwlabel and bweuler) it's giving me 6. It is probably because of the overlapping edges of the objects on the leftmost side?

How can I separate these overlaps so I can get an accurate count of 7?

Thank you!

Answer 1

[Arturo Jr. Ongkeko]

I was able to figure it out after some time. Here's the code.

%% COUNT BIG CELLS

%convert image to black and white
img_bw=im2bw(img);
figure,imshow(img_bw)

%background will be black and object will be white
inv_img=imcomplement(img_bw);
figure,imshow(inv_img)

%use imfill function to fill the holes in the object
img2=imfill(inv_img,'holes');
figure,imshow(img2)

%remove small objects containing fewer than 1000 pixels in the binary image 
%using bwareaopen
img3=bwareaopen(img2,1000);
figure,imshow(img3)

%use bwmorph erode function to further enhance the objects 
img4=bwmorph(img3,'erode');
figure,imshow(img4)

%conduct another removal of small objects containing fewer than 500 pixels
%in a binary image
img5=bwareaopen(img4,500);
figure,imshow(img5)

%use the bwlabel function to count the objects
[labeledImage,numberofObject]=bwlabel(img5); %count: 7

Answer 2

[DGM]

This works, but I had trouble trying to get watershed() to work without a little tweaking.

inpict = im2double(imread('cells.jpg'));

% try to extract the more bluish purple part
tol = [0.15 0.2]; % box width
clim = 0.615+[-1 1]*tol(1); % selected color point in C,H
hlim = 0.863+[-1 1]*tol(2);
cmap = mono(inpict,'clch');
hmap = mono(inpict,'hlch');
interior = cmap>clim(1) & cmap<clim(2) & hmap>hlim(1) & hmap<hlim(2);
% this is similar to bwareaopen(), but it does opening and/or closing
interior = despeckle(interior,20); 

d = bwdist(~interior);
d(d<(max(d(:))/2)) = 0; % had to give watershed a little help
L = watershed(~d);
L(~interior) = 0; % this is the label array

max(L) % this is the number of labeled objects

imshow2(simnorm(L))

This uses MIMT tools just because I'm lazy and can't be bothered to do it all in base/IPT tools

https://www.mathworks.com/matlabcentral/fileexchange/53786-image-manipulation-toolbox

[Arturo Jr. Ongkeko]

Thanks, DGM! Appreciate this! I am graduate student though and we are expected to use the IPT. Also we were not taught how to use image manipulation toolbox yet. I will defintely try your appoach for sure in the future. Thanks again very much!

[DGM]

Well it's not like MIMT is a thing anyone should be taught. It's good you solved it though.