glasbeyish

This is a library that implements a method described in the paper "Colour Displays for Categorical Images" by Glasbey et al. for finding optimal distinct color palettes of a specified size.

Unlike the glasbey library, this library implements the simulated annealing method described in the paper rather, than the iterative method. The simulated annealing method can often give better color palettes at the expense of longer run times.

Examples

These examples can also be found in the examples.ipynb Jupyter notebook.

import numpy as np
from nice_colorsys import rgb
from glasbeyish import glasbey
from matplotlib.colors import ListedColormap

Generate a palette with 2 colors

cm = ListedColormap([c.to_rgb() for c in glasbey(2)])
cm

Generate a palette with 12 colors

cm = ListedColormap([c.to_rgb() for c in glasbey(12)])
cm

Use a seed for reproducibility

cm = ListedColormap([c.to_rgb() for c in glasbey(12, seed=485)])
cm

cm = ListedColormap([c.to_rgb() for c in glasbey(12, seed=485)])
cm

Limit the colors that can be generated

Here, we'll try to make a bluish color palette by discarding colors too far from blue.

def color_dist(a, b):
    return np.linalg.norm(np.array(a.to_cieluv()) - np.array(b.to_cieluv()))

cm = ListedColormap([c.to_rgb() for c in glasbey(12, seed=485, limit=lambda x: color_dist(x, rgb(0, 0, 1)) < 100)])
cm

Use a custom random color generator

In this example, we'll generate random colors with only blue components in the RGB color space. glasbey expects the generated colors to be in the CIELUV color space and won't automatically convert the colors from RGB to CIELUV for us, so we have to do that ourselves.

rng = np.random.default_rng()

def random_color(rng):
    return rgb(0, 0, rng.random()).to_cieluv()
    
cm = ListedColormap([c.to_rgb().safe() for c in glasbey(12, seed=485, random_color=random_color)])
cm

Use a custom color distance

As an example, this code shows using a custom color distance to generate a palette for people with protanopia, a form of color blindness.

Disclaimer: This has not been tested with color blind users and might not even correctly assess color differences as seen by protanopes—it's for demonstration purposes only!

hpe = np.array([
    [0.4002, 0.7076, -0.0808],
    [-0.2263, 1.1653, 0.0457],
    [0,      0,      0.9182]
])
hpe_inv = np.linalg.inv(hpe)
lms = colorspace("lms", ["long", "medium", "short"])
register_space(lms, *derived_rgb_functions(ciexyz, lms, lambda x: hpe_inv @ x, lambda x: hpe @ x))

# Protanopia matrix from https://ixora.io/projects/colorblindness/color-blindness-simulation-research/
protan_matrix = np.array([
    [0, 1.05118294, -0.05116099],
    [0, 1,          0          ],
    [0, 0,          1          ]
])
def protan(x):
    return np.array(lms(*(protan_matrix @ cieluv(*x).to_lms())).to_cieluv())

def protan_dist(a, b):
    a = protan(a)
    b = protan(b)
    return np.linalg.norm(b - a)

cm = ListedColormap([c.to_rgb().safe() for c in glasbey(8, seed=485, color_dist=protan_dist)])
cm