📘

K-Means Color Quantization

Implement K-means clustering for image color quantization using NumPy.

Color quantization reduces the number of distinct colors in an image, used for compression and artistic effects.

Algorithm:

Reshape image from (H, W, 3) to (H×W, 3) - each row is an RGB pixel
Initialize K cluster centers (random pixels or k-means++)
Iterate until convergence:
- Assign each pixel to nearest center (Euclidean distance)
- Update centers to mean of assigned pixels
Replace each pixel with its cluster center

Distance metric: $d(p, c) = \sqrt{(r_p-r_c)^2 + (g_p-g_c)^2 + (b_p-b_c)^2}$

Input:

image = [[[255,0,0], [0,255,0]],
         [[0,0,255], [255,255,255]]]
k = 2

Output:

Colors quantized to 2 clusters

Reasoning:

Step 1: Reshape Pixels: [[255,0,0], [0,255,0], [0,0,255], [255,255,255]]

Step 2: Initialize centers Random selection: e.g., [255,0,0] and [255,255,255]

Step 3: Iterate Assign pixels to nearest center, update centers...

After convergence, each pixel is replaced by its cluster center color.

With k=2 on this image, typical result groups similar colors together.

Editor

Python 3.13.1

0/0

Run code to see test results.

📘