kmeans - stylized images using image segmentation and pixelation

This project implements an algorithm for stylizing images that clusters pixels by color, then maps each cluster to another color. Optionally, the image can be downsampled or have noise added prior to mapping colors, which produces a pixelated or grainy effect when reupscaled.

Before	After

Photograph taken by Daniel Mennerich (Flickr).

Before

After

Photograph taken by Macaolards (Wikipedia).

setup

It is recommended to use a virtual environment to manage dependencies.

python -m venv venv
source venv/bin/activate
python -m pip install -r requirements.txt

usage

Using the GUI application allows tuning the pipeline parameters and previewing the effects in real-time.

python gui.py

The CLI application has the same features as the GUI application. Run using the following arguments:

python cli.py <filename> <palette> <pixelsize> <space> <noisesize>

arguments

argument	description	default value
filename	str: The filename of the image to run the algorithm on.	None (required)
palette	int | str: Either the number of colors to use for an auto-generated palette, or the name of a builtin palette (see palettes).	16
pixelsize	int | None: If int, the size of each pixel after reupsampling. If None, no downsampling is performed.	None
space	str: The color space to run the pipeline in. One of rgb, linear, or oklab.	rgb
noisesize	number: The amount of noize to add to the image, measured in standard deviations. Effect varies depending on space.	0.0

palettes

palette	description
latte	Catppuccin Latte
macchiato	Catppuccin Macchiato

acknowledgments

• Catppuccin