k means clustering, error

im = imread('hestain.png');
im=rgb2gray(im) //if you only want grayscale intensities
[idx centroids]=kmeans(double(im(:)),4);

??? Error using ==> kmeans>batchUpdate at 436
Empty cluster created at iteration 1.

Error in ==> kmeans at 337
        converged = batchUpdate();

Error in ==> Untitled2 at 5
[idx centroids]=kmeans(double(im(:)),4);

Does anyone help ?

Thanks.

Answer 1

[Walter Roberson]

Look at the documentation page, at the 'emptyaction' option.

Answer 2

[Image Analyst]

Evidently you've adapted the demo at this web page to your image incorrectly. You don't have clusters because you don't have a second axis. You just passed in a 1D vector. What is the other axis that you think you have that forms clusters?

[Tomas]

i want coordinates position points in image and color

[Image Analyst]

That does not make sense. You need to study up some more on what kmeans means.

[Tomas]

I understand k means method. I don't understand image proccesing and computer vision.

give examples about what I need

in binary images

I,map]=imread('test3','bmp');
I = ~I;

imshow(I,map);

[m n]=size(I)

P = [];
for i=1:m    
    for j=1:n        
        if I(i,j)==1
            P = [P ; i j];        
        end
    end
end

size(P)
MON=P;

[IDX,ctrs] = kmeans(MON,3)
clusteredImage = zeros(size(I));
clusteredImage(sub2ind(size(I) , P(:,1) , P(:,2)))=IDX;
imshow(label2rgb(clusteredImage))

my output:

I need this also with grayscale and RGB image.

[Image Analyst]

You can do simple color classification by simple thresholding in RGB color space with that image. No kmeans clustering is needed. See my rgb color classification demo in my File Exchange: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862 For example if you want a map (binary image) of where all the pure blue, cyan, or yellow pixels are, you simply do this:

% Extract the individual red, green, and blue color channels.
redChannel = rgbImage(:, :, 1);
greenChannel = rgbImage(:, :, 2);
blueChannel = rgbImage(:, :, 3);

pureYellowPixels = redChannel == 255 & greenChannel == 255 & blueChannel == 0;
pureCyanPixels = redChannel == 0 & greenChannel == 255 & blueChannel == 255;
pureBluePixels = redChannel == 0 & greenChannel == 0 & blueChannel == 255;

If you want a N by 2 listing of all the (row, column) pairs where the pixels lie, just do this:

[rowsY, columnsY] = find(pureYellowPixels);
[rowsC, columnsC] = find(pureCyanPixels);
[rowsB, columnsB] = find(pureBluePixels);

But that kind of thing is rarely needed.

[Tomas]

ok, Thank you for your time and help.

I have to use clustering methods.

My graduation theses is about cluster analysis.

[Image Analyst]

Then you must know more about it than me. So again I ask you what are the two measures, what are the two axes that something is going to be plotted against, if you're not going to use the locations (rows, columns) like I derived for you? And if you do use those, then there's no need to classify because the classification was done first in order to get the coordinates.