how to set the centroid in kmean clustering so that the cluster values do not change

hi, i am learning how to segment colors by using kmean clustering just like the example in matlab 2015a. but each time i run the codes, the colors that i want are in different clusters. for example, for the first run,it will display that yellow is in cluster 1 and blue is in cluster 2. but when i run it again, they will switch to different cluster. how to make the yellow and blue is in specific clusters even if i run it again and again? please help me. thanks in advance

    [FileName,PathName] = uigetfile('*.jpg','Select the MATLAB code file');
 he1= imread(FileName);
 cform = makecform('srgb2lab');
 lab_he = applycform(he1,cform);
 figure (2)
 imshow (lab_he)
 ab = double(lab_he(:,:,2:3));
 nrows = size(ab,1);
 ncols = size(ab,2);
 ab = reshape(ab,nrows*ncols,2);
 nColors = 3;
 % repeat the clustering 3 times to avoid local minima
 [cluster_idx, cluster_center] = kmeans(ab,nColors,'distance','sqEuclidean',                          ...
                              'Replicates',3);
 pixel_labels = reshape(cluster_idx,nrows,ncols);
 figure (3)
 imshow(pixel_labels,[]), title('image labeled by cluster index');
 segmented_images = cell(1,3);
 rgb_label = repmat(pixel_labels,[1 1 3]);
 for k = 1:nColors
     color = he1;
     color(rgb_label ~= k) = 0;
     segmented_images{k} = color;
 end
 figure (4)
 imshow(segmented_images{1}), title('objects in cluster 1');
 figure (5)
 imshow(segmented_images{2}), title('objects in cluster 2');
 figure (6)
 imshow(segmented_images{3}), title('objects in cluster 3');

Answer 1

[Image Analyst]

If you have reference CIE LAB values, then simply compute the delta E of each pixel in the image to each of your reference colors. Then I think you can stack all the delta E images into a 3D array and use min() to get the class (ref color that is closest) for each pixel. For example, if you have 4 reference colors, then you will have 4 delta E images, and you can do (I think - untested)

deltaE1 = sqrt((lImage - refL1).^2 + (aImage - refA1).^2 + (bImage - refB1).^2);
deltaE2 = sqrt((lImage - refL2).^2 + (aImage - refA2).^2 + (bImage - refB2).^2);
deltaE3 = sqrt((lImage - refL3).^2 + (aImage - refA3).^2 + (bImage - refB3).^2);
deltaE4 = sqrt((lImage - refL4).^2 + (aImage - refA4).^2 + (bImage - refB4).^2);
deltaE = cat(3, deltaE1, deltaE2, deltaE3, deltaE4);
[~, classifiedImage] = min(deltaE, 3);

[Mohammad Farhad Aryan]

How do you apply your answer in original above code?

[Image Analyst]

Mohammad, I'm attaching two of my delta E demos. If neither one does what you want, then write back and say what you want differently.