How can PIL and numpy be employed to quantize an image?

edit

Quantization is the process of reducing the number of distinct colors in an image. Here's how you can use PIL and numpy to quantize an image:

1. Read the image using PIL's Image module

from PIL import Image
img = Image.open("image.jpg")

1. Convert the image to a numpy array

import numpy as np
img_arr = np.array(img)

1. Flatten the numpy array to a 2-dimensional array (pixels x 3 RGB channels)

img_arr_flat = img_arr.reshape(-1, 3)

1. Perform k-means clustering on the flattened array, where k is the desired number of colors

from sklearn.cluster import KMeans
kmeans = KMeans(n_clusters=16) # 16 colors
kmeans.fit(img_arr_flat)

1. Get the centroid values of the clusters

centroid_vals = kmeans.cluster_centers_.astype(int)

1. Replace the pixel values in the original numpy array with the nearest centroid pixel value

centroid_idx = kmeans.predict(img_arr_flat)
quantized_arr_flat = centroid_vals[centroid_idx]
quantized_arr = quantized_arr_flat.reshape(img_arr.shape)

1. Convert the quantized numpy array back to a PIL Image object

quantized_img = Image.fromarray(np.uint8(quantized_arr))

1. Save the quantized image

quantized_img.save("quantized_image.jpg")

This will produce an image with the same dimensions as the original, but with a reduced color palette of 16 colors. You can adjust the number of colors by changing the n_clusters parameter in step 4.