Use k-means clustering to identify primary colors of an input image. Can specify --n-colors=INT. Also, --out-pallete=FILE will create a pallete of colors identified.

average_image.py

import argparse
import numpy as np
from scipy import ndimage
from scipy.cluster.vq import kmeans
from scipy.misc import imsave


def rgb2hex(r, g, b):
    return '#{:02x}{:02x}{:02x}'.format(r, g, b)


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument("image", help="Path to image file to process.")
    parser.add_argument("--n-colors", "-n", default=3, type=int, help="Number of colors to pick.")
    parser.add_argument("--out-pallete", default=None, help="Make an optional pallete image.")

    args = parser.parse_args()

    img = ndimage.io.imread(args.image)
    n_pixels = img.shape[0] * img.shape[1]
    pixels = img[:, :, 0:3].reshape(n_pixels, 3)

    result = kmeans(pixels, args.n_colors)

    if args.out_pallete:
        p = np.zeros((100 * args.n_colors, 100, 4))

    for ix, color in enumerate(result[0]):
        r, g, b = color
        hexstr = rgb2hex(r, g, b)
        print("(%d, %d, %d)\t(%s)" % (r, g, b, hexstr))

        if args.out_pallete:
            p[(ix * 100):((ix + 1)*100), :, 0:3] = color
            p[(ix * 100):((ix + 1)*100), :, 3] = 255

    if args.out_pallete:
        imsave(args.out_pallete, p)