kmeans

python wrapper for a basic c implementation of the k-means algorithm.

Please review the limitations before using in any capacity where strict accuracy is required. There is no overflow detection, and negatives are not supported. tuple values cannot exceed 255.

Installation

pip install kmeans

Usage

import kmeans
means = kmeans.kmeans(points, k)

points should be a list of tuples of the form (data, weight) where data is a list with length 3.

For example, finding four mean colors for a group of pixels:

pixels = [
    [(15, 20, 25), 1],  # [(r,g,b), count]
    [(17, 31, 92), 5],
    # ... Lots more ...
]

centers = kmeans.kmeans(pixels, 4)

In this case, the weights passed in may be the frequency of the pixels occuring in an image, or some preference to pull the means towards a color.

Limitations

All values must be non-negative integers, with the following restrictions:

r, g, b        [0, 255]        (uint8_t)
count          [0, 4294967295] (uint32_t)
maximum points 4294967296      (uint32_t)
maximum means  256             (uint8_t)
max iterations 65536           (uint16_t)
max tolerance  65536           (uint16_t)

Inspiration

http://charlesleifer.com/blog/using-python-to-generate-awesome-linux-desktop-themes/

I wanted to apply the implementation there to images much larger than 200x200. Running a 4k x 3k image was approaching 60 seconds on a nice computer, so I decided to rewrite the kmeans implementation in c.

Contributors

• chinatsu - 32bit compatibility