def create_clustering_mask( em, image ):
    '''
    Create clustering mask
    '''
    rows, cols, _ = image.shape
    samples       = convert_to_samples( image )
    mask          = []
    pixels_list   = [None for x in range(255255)] # smaller foot print than dict

    print (rows * cols)

    for index in range( rows * cols ):
        sample = samples[index:index+1, :]

        # rather than using more expensive hash, we know that
        # S and V ranges between 0 - 255, so by using this
        # we could uniquely store values between 000000 to 255255
        key = int(sample[0][0] * 1000 + sample[0][1])
        cluster = pixels_list[key]

        # Cache miss
        if cluster is None:
            cluster, prob = em.predict( sample )
            pixels_list[key] = cluster

        mask.append( cluster )

    return np.array( mask ).reshape( rows, cols, 1 ).astype( 'uint8' )