Remove single pixel width connecting lines between objects

Hi all,

How can I remove/disconnect the connecting lines between objects while keeping the boundaries of my objects intact?

[KALYAN ACHARJYA]

As per previous discussion, have you tried with Morphological operations?

[Haider Ali]

@KAYLAN, yes I have tried some morphological operations (erosion and closing) but erosion erodes the cells' boundaries which is undesirable. Maybe I am not choosing the structuring element wisely. Can you please suggest something?

Thanks

[Matt J]

I recommend that you attach your image in a .mat file to speed things up.

[Haider Ali]

I have attached the image in .mat file.

[Matt J]

An intriguing problem.

[darova]

=)

Answer 1

[darova]

Here is an idea: find large areas and merge them to remove walls

I = imread('image.png');
I1 = im2bw(I);
% I2 = bwmorph(~I1,'thin',3);
[I3,n] = bwlabel(I1,8);                 % label image
p = regionprops(I3,'Area');             % find area of each region
ar = cat(1,p.Area);
ind = find( 400 < ar & ar < 5000);      % find interested areas
I4 = I3*0;
for i = 1:length(ind)
    ix = find( I3==ind(i) );            % find pixels of interested areas
    I4(ix) = 1;
end
I5 = imdilate(I4,ones(3));              % merge neigbouring areas
I6 = imdilate(~I5,ones(3));             % return original size of areas

II = cat(3,I1,~I6,I1*0);
imshow(II)
figure
imshow(I1 | ~I6)

The result: Green lines/walls removed, Yellow areas detected

[Haider Ali]

@darova, interesting sloution.

But I am afraid it does not solve my actual problem.

The image that I have provided is actaully obtained after applying watershed on binarized cell image. My goal is to segment the cells and calculate their properties.

The probelm I am facing is that after binarization (imbinarize), some of the cells' boundaries are not closed> Therefore, I applied the waterhsed transform. This improves the cell segmentation but also introduces extra ridges (connecting lines) and these extra ridges create false positives.

So there are couple of things that I can do:

1. Improve my binarization/thresholding algorithm to create strong edged cell boundaries.
2. Perform some pre-processing before watershedding to join broken cells' boundaries (please see this question too Edge linking and hole filling in binarized cell image).
3. Remove the newly created connecting lines introduced by watershed transform.

What would the best approach in your opinion?

I am attaching some figures too.

In watershedimage.fig, I have circled 2 false positives which were introduced only because of watwershed transform.

Please let me know if any pther explanation of the problem is required.

Thanks.

[darova]

I'm afraid that i can't help. All i can advise is binarizing image and manipulating with graythersh