Reducing variance of scipy.cluster to find dominant colours

Question

import numpy, scipy.cluster, collections
from PIL import Image, ImageDraw

def dominant_colors(image): 
    """
    Input: PIL image
    Output: A lis