Median cut vs. k-means clustering