Applying Voronoi tessellation to images

In [1]:

    

from scipy.ndimage import imread
from scipy.misc import imsave, imresize

import pylab as plt
import numpy as np
from scipy import spatial
from sklearn import neighbors

from collections import defaultdict

%matplotlib inline
plt.rcParams['figure.figsize'] = (16.0, 9.0)


    

In [2]:

    

def print_status(msg):
    from IPython.display import clear_output
    import sys
    clear_output()
    print msg
    sys.stdout.flush()


    

In [3]:

    

I = imread('cat.jpg')
I.shape


    

    
Out[3]:





(1024, 1024, 3)

In [4]:

    

plt.imshow(I)


    

    
Out[4]:





<matplotlib.image.AxesImage at 0x115599b10>






    

In [5]:

    

nseeds = 5000


    

In [6]:

    

seeds = np.vstack([np.random.uniform(size=nseeds) * I.shape[0],
                   np.random.uniform(size=nseeds) * I.shape[1]]).T


    

In [ ]:

    

plt.imshow(I)
plt.plot(seeds[:,1], seeds[:,0], 'g.')


    

    
Out[ ]:





[<matplotlib.lines.Line2D at 0x1155c6910>]






    

In [ ]:

    

def voronoify_np():
    cells = defaultdict(list)

    for r in range(I.shape[0]):
        for c in range(I.shape[1]):
            label = np.argmin(np.sum((seeds - [r,c])**2, axis=1))
            cells[label].append([r,c])

    I2 = I.copy()
    for idx in cells.values():
        idx = np.array(idx)
        mean_col = I[idx[:,0], idx[:,1]].mean(axis=0)
        I2[idx[:,0], idx[:,1]] = mean_col

    plt.imshow(I2)
    
    
%time voronoify_np()


    

In [ ]:

    

def voronoify_sklearn(I, seeds):
    tree_sklearn = neighbors.KDTree(seeds)
    pixels = ([(r,c) for r in range(I.shape[0]) for c in range(I.shape[1])])
    d, pos = tree_sklearn.query(pixels)

    cells = defaultdict(list)
    for i in range(len(pos)):
        cells[pos[i][0]].append(pixels[i])
    
    I2 = I.copy()
    for idx in cells.values():
        idx = np.array(idx)
        mean_col = I[idx[:,0], idx[:,1]].mean(axis=0)
        I2[idx[:,0], idx[:,1]] = mean_col
    
    plt.imshow(I2)
    
    
%time voronoify_sklearn(I, seeds)


    

In [ ]: