Babak Taati

This is a really nice function.

As Fen Xie pointed out, it sometimes returns empty clusters (esp if you set k, the desired number of clusters too high compare to the size of you data).

The solution is easy though. You only need two extra lines of code to get rid of the empty clusters:

% 1- dicsard empty clusters
centers = centers(:, unique(labels));

% 2- re-assign labels
[~,labels] = max(bsxfun(@minus,centers'*X,0.5*sum(centers.^2)'));