k means clustering shows only blank image if i loop it k times

If i loop for i = 1:2 i get clustering otherwise i just get a blank image. any idea why?

im = imread("irobot.jpg");
im_as_col = double(im(:));
cluster_membs = kmeans(im_as_col, 3)

cluster_membs = 6912000×1

2 2 2 2 2 2 2 2 2 2

labelim = zeros(size(im));
for i=1:3
    inds = find(cluster_membs==i);
    meanval = mean(im_as_col(inds));
    labelim(inds) = meanval;
end
imshow(im)

imshow(labelim);

[KSSV]

Error says there is no image in the path, did you specify the correct path of the image or is image present in the present working directory?

[Anirudh Kochhar]

hey sorry i forgot to add the image to the question. now that it is attached it should be easier to see. I dont get to see the k mean clustered image for some reason. it is a white image.

Answer 1

[yanqi liu]

clc; clear all; close all;
im = imread("https://ww2.mathworks.cn/matlabcentral/answers/uploaded_files/774998/irobot.jpg");
im_as_col = double(im(:));
cluster_membs = kmeans(im_as_col, 3);
labelim = zeros(size(im));
for i=1:3
    inds = find(cluster_membs==i);
    meanval = mean(im_as_col(inds));
    labelim(inds) = meanval;
end
figure; imshow(im,[])

figure; imshow(mat2gray(labelim),[]);

[Anirudh Kochhar]

Interesting solution. Could you give me more details over why this works?

[Image Analyst]

@Anirudh Kochhar I'm not sure it does work. You wanted 3 classes and that's what I gave you in my answer below. This answer's final image looks like it shows 5 classes: blue, green, cyan, white, and gray classes, though kmeans does return 3. And the classes are not based on colors since all the gray levels are lumped together with this line:

im_as_col = double(im(:));

so you won't find unique/similar colors as in my answer. A color like pure red [1,0,0] would show up as the same class as pure green [0,1,0] or pure blue [0,0,1] because all the gray levels were put into a single column.

But I'm not sure why you wanted only 3 color classes since that image has blue, green, cyan, and a wide variety of grays. Why did you pick 3? This is not an image that has 3 clusters. I mean, just look at the gamut:

colorcloud(im);

Do you see 3 natural clusters there? No.

[yanqi liu]

yes，sir，its make 3 color class，and use label to display different gray level，that can be ref label2rgb

Answer 2

[Image Analyst]

See attached demos.

[Image Analyst]

If you want kmeans for a hyperspectral image, see attached.

Answer 3

[Image Analyst]

For what it's worth, attached is my Color Classifier that is based on Discrminant Analysis instead of kmeans. Basically you draw regions that are representative of the colors you want to have as your classes. Then it classifies all the pixels in the image into one of those classes.

Also attaching a KNN classifier demo.

Adapt as needed.