Image quantization

In previous sections, we used clustering to explore the structure of a dataset. Now let's apply it to a different problem. Image quantization is a lossy compression method that replaces a range of similar colors in an image with a single color. Quantization reduces the size of the image file since fewer bits are required to represent the colors. In the following example, we will use clustering to discover a compressed palette for an image that contains its most important colors. We will then rebuild the image using the compressed palette. First we read and flatten the image:

# In[1]:
import numpy as np
import matplotlib.pyplot as plt
from sklearn.cluster import KMeans
from sklearn.utils import shuffle
from PIL import Image

original_img = np.array(Image.open('tree.jpg'), dtype=np.float64) /
  255
original_dimensions = tuple(original_img.shape)
width, height, depth = tuple(original_img.shape)
image_flattened = np.reshape(original_img, (width * height, depth))

We then use...