A color palette extractor written in Python using KMeans clustering.
Working with computer graphics and visualizations, one often needs a way of specifying a set of colors with a certain visual appeal. Such a set of colors is often called a color palette. The aim of this library is to easily extract a set of colors from a supplied image, with support for the various color modes (RGB, RGBa, HSV, etc). Dabbling in generative art, the need often arises for being able to pick colors at random from a palette. Pylette supports this, both picking colors uniformly, but also using the color frequency from the original image as probabilities.
Other color palette related Python-libraries:
- Color Thief: Extraction of color palettes using the median cut algorithm.
- Palettable: Generation of matplotlib compatible color schemes
- Colorgram: Extraction of colors from images (similar to the intended use of this library), however, I was unable to install this.
Pylette is available in the python package index (PyPi), and can be installed using
pip install Pylette
Palette object is created by calling the
from Pylette import extract_colors palette = extract_colors('image.jpg', palette_size=10, resize=True) palette = extract_colors('image.jpg', palette_size=10, resize=True, mode='MC', sort_mode='luminance')
This yields a palette of ten colors, and the
resize flag tells Pylette to resize the image to a more manageable size before
beginning color extraction. This significantly speeds up the extraction, but reduces the faithfulness of the color palette.
One can choose between color quantization using K-Means (default) or Median-Cut algorithms, by setting in the
mode-parameter. One can also specify to alternatively sort the color palette by the luminance (percieved brightness).
The palette object supports indexing and iteration, and the colors are sorted from highest to lowest frequency by default. E.g, the following snippet will fetch the most common, and least common color in the picture if the palette was sorted by frequency, or the darkest to lightest color if sorted by luminance:
most_common_color = palette least_common_color = palette[-1] three_most_common_colors = palette[:3]
As seen, slicing is also supported.
The Palette object contains a list of Color objects, which contains a representation of the color in various color modes, with RGB being the default. Accessing the color attributes is easy:
color = palette print(color.rgb) print(color.hls) print(color.hsv)
To display the extracted color palette, simply call the
display-method, which optionally takes a flag for saving the palette to an image file.
The palette can be dumped to a CSV-file as well, where each row represents the RGB values and the corresponding color frequency (optional).
palette.display(save_to_file=False) palette.to_csv(filename='color_palette.csv', frequency=True)
In order to pick colors from the palette at random, Pylette offers the
random_color-method, which supports both drawing
uniformly, and from the original color distribution, given by the frequencies of the extracted colors:
random_color = palette.random_color(N=1, mode='uniform') random_colors = palette.random_color(N=100, mode='frequency')
A selection of example palettes. Each palette is sorted by luminance (percieved brightness). The top row corresponds to extraction using K-Means, and the bottom row corresponds to Median-Cut extraction.
|Original Image||Extracted Palette|
Under the hood
Currently, Pylette uses KMeans for the color quantization. There are plans for implementing other color quantization schemes, like:
- Median-cut [Implemented]
- Modified minmax
The article Improving the Performance of K-Means for Color Quantization gives a nice overview of available methods.
Any feedback and suggestions is much appreciated. This is very much a work in progress.