How to cluster rows of a matrix and label them

I have to convexify Kmeans clustering. So I'll be using some modified functions. I want to know how do I label my modified image matrix(80000x3 double) and form clusters and initialise the cluster centres by taking the mean of the cluster formed(I have to make sure that the initailised cluster centres lie within the data points) . (Generally, cluster centres are generated randomly initially.) Please help with the code.

I = im2double(imread('hest.png'));
F =reshape(I,size(I,1)*size(I,2),3);
K = 5; % # of cluster
U = F( ceil(rand(K,1)*size(F,1)) ,:);% random cluster centres generated 

Answer 1

[Image Analyst]

See my attached demo to do kmeans clustering on an RGB image. Adapt as needed.

Generally kmeans gives a crummy segmentation with color images. You might try discriminant analysis instead (demo also attached). It seems to do better.