Homework 3 Problem 5



In [1]:

    
import numpy as np
import matplotlib.pyplot as plt
import kmeans

KCENTROIDS = 16
MAX_ITERS = 200
EPSILON = 1e-7

Part 5.a.

Here we load the large mandrill image and display it.



In [2]:

    
Ilarge = plt.imread('mandrill-large.tiff')
Ilarge = np.uint8(np.round(Ilarge))



In [3]:

    
plt.imshow(Ilarge)
plt.show()

Part 5.b.

Here we load the small mandrill image and display it.



In [3]:

    
Ismall = plt.imread('mandrill-small.tiff')
Ismall = np.uint8(np.round(Ismall))



In [4]:

    
plt.imshow(Ismall)
plt.show()

Now apply the K-Means algorithm with k centroids to the small image.



In [5]:

    
clusters_small, centroids_small = kmeans.kmeans(Ismall, kcentroids=KCENTROIDS, epsilon=EPSILON, max_iterations=200)

The color palette obtained by running K-Means is the following.



In [6]:

    
plt.imshow(np.array([centroids_small]))
plt.show()



In [7]:

    
new_image_small = kmeans.make_image(clusters_small, centroids_small)



In [8]:

    
plt.imshow(new_image_small)
plt.show()

Part 5.c.

Do the same as above for the large mandrill image.



In [10]:

    
clusters_large, centroids_large = kmeans.kmeans(Ilarge, kcentroids=KCENTROIDS, epsilon=EPSILON, max_iterations=MAX_ITERS)



In [11]:

    
new_image_large = kmeans.make_image(clusters_large, centroids_large)



In [12]:

    
plt.imshow(new_image_large)
plt.show()

Part 5.d.

The original image requires 24 bits per pixel to represent the image. Compressing the representation to 16 colors using K-Means clustering requires log2(16) = 4 bits per pixel. The resulting compression factor is 24 / 4 = 6.



In [ ]: