This is an implementation of the k-means clustering algorithm. Image pixels are represented as 4-dimensional vectors (r,g,b,a) and each pixel is ultimately assigned to one of k clusters. Because the cluster centroids are initialized on random pixels, the final image is (usually) different on each run.
After cloning the repo, npm install. You may need to install ImageMagick, which is easiest done with Homebrew. Drop a .jpg image into the directory, and configure the variables jpeg_name, NUM_ITERATIONS, and k in k-means.js accordingly. Then run node k-means.js to transform your image!
An interesting pattern emerged while generating k-means'd versions of my profile picture. In 10 runs of the algorithm, with 10 iterations each and K=3, each run appeared to converge at one of only two local optima: .png "Title") .png "Title")
(See the images/10_3/ directory for all 10 runs).
The code and this README are both works-in-progress. Feel free to contact me with suggestions!