Voronoiify an image

voronoi.py

"""
This divides the images in small patches by means of minimal distance to random points.
It then computes the average color of each patch, and gives the whole patch that color.
This code is NOT fast, O(N^2) for N number of pixels (~1min). It's pretty short and easy though.
"""

from random import randint, seed
from time import time
from matplotlib.pyplot import subplots, show
from numpy import zeros, array, int32
from scipy.misc import imread


def voronoi(img, centers):
	partof = zeros((img.shape[0], img.shape[1], 2,))
	sums = dict((center, array([0, 0, 0, 0], dtype=int32)) for center in centers)
	for px in range(img.shape[0]):
		for py in range(img.shape[1]):
			min_xy = (-1, -1)
			min_dist = 1e100
			for cx, cy in centers:
				dist = (cx - px) ** 2 + (cy - py) ** 2
				if dist < min_dist:
					min_xy = cx, cy
					min_dist = dist
			partof[px, py, :] = min_xy
			pix = img[min_xy]
			sums[min_xy] += array([pix[0], pix[1], pix[2], 1])
	avgs = dict((k, [s[0] // s[3], s[1] // s[3], s[2] // s[3]]) for (k, s) in sums.items())
	for px in range(img.shape[0]):
		for py in range(img.shape[1]):
			img[px, py, :] = avgs[tuple(partof[px, py, :])]
	return img


if __name__ == '__main__':
	img = imread('your_image_name_here.png')
	fig, (before, after) = subplots(2, 1, figsize=(8, 12), tight_layout=True)
	before.imshow(img)
	centercnt = int((img.shape[0] * img.shape[1]) / 200)
	seed(42424242)
	centers = set()
	while len(centers) < centercnt:
		centers.add((randint(0, img.shape[0] - 1), randint(0, img.shape[1] - 1)))
	print('voronoi for {} centers on {}x{} img'.format(centercnt, img.shape[0], img.shape[1]))
	t0 = time()
	voro = voronoi(img, centers)
	print('voronoi took {:.3f}s'.format(time() - t0))
	after.imshow(voro)
	show()

Inspired by https://github.com/Grabot/ArtProject