Image segmentation using k means clustering

Hello, I have a question and I appreciate your help. I don't know how to use a kmeans clustering results in image segmentation. I have an RGB image of a tissue which has 5 colors for 5 biomarkers and I need to do k means clustering to segment every color in a cluster. Thank you so much for your help.

Answer 1

[Image Analyst]

I'm attaching a color kmeans demo that doesn't require a recent version of the stats toolbox or image processing toolbox.

Personally I don't think kmeans works that great (perhaps because it's untrained), as you'll probably see, and you might try a trained discriminant analysis instead. I attach demos for both.

[Penny13]

Thank you so much Image Analyst. I will try both to see what would be the results.

Answer 2

[Walter Roberson]

If you have 5 biomarkers then you would need to segment to a minimum of 6 clusters: one for each marker and one for tissue that is not one of the biomarkers. kmeans always assigns a cluster to every point, so if you had a point that was not one of the 5 colors and you asked to cluster it, then it would assign it to one of the five anyhow.

Note that the index values returned by kmeans are not in any pre-set order. You cannot assume that index 1 is associated with biomarker color #1. It is common for clusters to change effective identities during processing, so even if you had specified initial cluster centers in the options you passed to kmeans, you should not assume that the numbering of the output will be the same as the order of the initial cluster centers.

Remember too that a cluster centroid can be outside of data that belongs to that cluster. For example consider a perfect unfilled semi-circle of points: its centroid is the centre of the circle but no data points are at that center. You therefore cannot just index the centroid location into the image and look at the pixel value there to figure out what is happening.

[Penny13]

Thank you so much for your answer.

clear all;
close all;

   he = imread('48rf.jpg');
  % he = imread('11111.jpg');
  % he = imread('tt.jpeg');
  figure;
  imshow(he), title('H&E image');
   cform = makecform('srgb2lab');
  lab_he = applycform(he,cform);
  ab = double(lab_he(:,:,2:3));
  nrows = size(ab,1);
  ncols = size(ab,2);
  ab = reshape(ab,nrows*ncols,2);

nColors = 5;
% repeat the clustering 5 times to avoid local minima
[cluster_idx, cluster_center] = kmeans(ab,nColors,'distance','sqEuclidean', ...
                                      'Replicates',5);
                                  pixel_labels = reshape(cluster_idx,nrows,ncols);
 figure;                                 
imshow(pixel_labels,[]), title('image labeled by cluster index');
segmented_images = cell(1,5);
rgb_label = repmat(pixel_labels,[1 1 3]);

for k = 1:nColors
    color = he;
    color(rgb_label ~= k) = 0;
    segmented_images{k} = color;
end
figure;
imshow(segmented_images{1}), title('objects in cluster 1');
figure;
imshow(segmented_images{2}), title('objects in cluster 2');
figure;
imshow(segmented_images{3}), title('objects in cluster 3');

figure;
imshow(segmented_images{4}), title('objects in cluster 4');

figure;
imshow(segmented_images{5}), title('objects in cluster 5');

Sir, I used this code from one MATLAB blogs on my image, but it keeps telling me that too many input arguments. Could you please help me what is the problem? My apologies if my question is basic.

[Walter Roberson]

Which MATLAB release are you using?

[Penny13]

MATLAB_R2017b

[Walter Roberson]

Could you post the entire error message ? And can you attach the .jpeg so we can test?

[Penny13]

Error using kmeans Thanks so much. I attached the image.

Too many input arguments. Error in C_B_Seg_K_Means_Clustering (line 19) [cluster_idx, cluster_center] = kmeans(ab,nColors,'distance','sqEuclidean', ...

[Walter Roberson]

No error for me here. Try

which -all kmeans

I suspect you are getting a third-party kmeans

[Penny13]

Thank you so much for your reply. I used the code from this webpage: https://www.mathworks.com/help/images/examples/color-based-segmentation-using-k-means-clustering.html

May I ask if you tried this code? Or another code? Thanks a lot for your help

[Walter Roberson]

I have not used that code; I do not work on image segmentation.

[yuvraj singh]

imsegkmeans doesn't work

[Walter Roberson]

imsegkmeans appears to have been introduced in R2018b in the Image Processing Toolbox. Are you using R2018b ?

[Penny13]

Yes.I work with MATLAB 2018 version.

[Image Analyst]

Poupack, I use this code and I get no such error. Please try it. If you get an error, copy ALL the red text and paste it back here:

clear all;
close all;
he = imread('peppers.png');
% he = imread('11111.jpg');
% he = imread('tt.jpeg');
subplot(3, 3, 1);
imshow(he), title('H&E image');
cform = makecform('srgb2lab');
lab_he = applycform(he,cform);
ab = double(lab_he(:,:,2:3));
nrows = size(ab,1);
ncols = size(ab,2);
ab = reshape(ab,nrows*ncols,2);
nColors = 5;
% repeat the clustering 5 times to avoid local minima
[cluster_idx, cluster_center] = kmeans(ab,nColors,'distance','sqEuclidean', ...
	'Replicates',5);
pixel_labels = reshape(cluster_idx,nrows,ncols);
subplot(3, 3, 2);
imshow(pixel_labels,[]), title('image labeled by cluster index');
segmented_images = cell(1,5);
rgb_label = repmat(pixel_labels,[1 1 3]);
for k = 1:nColors
	color = he;
	color(rgb_label ~= k) = 0;
	segmented_images{k} = color;
end
subplot(3, 3, 3);
imshow(segmented_images{1}), title('objects in cluster 1');
subplot(3, 3, 4);
imshow(segmented_images{2}), title('objects in cluster 2');
subplot(3, 3, 5);
imshow(segmented_images{3}), title('objects in cluster 3');
subplot(3, 3, 6);
imshow(segmented_images{4}), title('objects in cluster 4');
subplot(3, 3, 7);
imshow(segmented_images{5}), title('objects in cluster 5');

[SRUTHI BANDI]

Error in demo1 (line 16)

[cluster_idx, cluster_center] = kmeans(ab,nColors,'distance','sqEuclidean', ...

error comes like this

[Image Analyst]

An error like what? You forgot to give the actual error message.

I just copied and pasted and ran it again successfully.

Are you sure you have the Stats toolbox?