How to choose a good colormap

Damon McDougall sharing information about how to choose a good colormap.

Transcript

1. How to choose a good colour map Damon McDougall Institute for Computational and Engineering Sciences, UT Austin, USA 10th July 2014 1/19
2. Introduction Most of the content is taken from this excellent article: http://www.research.ibm.com/people/l/lloydt/color/color.HTM 2/19
3. Introduction • Data is a huge aspect of science • By and large we (scientists) treat data well. . . • . . . and we visualise it poorly. Why? • Colour maps • Data is of some field f : Ω ⊂ R2 → [0, 1] • A colour is assigned to the output of f (a scalar). Seems reasonable. • Colour map is a function g : [0, 1] → Ω ⊂ R3 • Mismatch in dimensions: R3 versus R • The point? Colour maps can be misleading. 3/19
4. Hating on the jet colour map What is this? 4/19
5. Hating on the jet colour map Did anybody see Florida? • Left: Linear interpolation in RGB space between red and blue. • Right: Changes in data are perceived as proportional changes in colour (subjective) • Right: Domain specific knowledge used to reveal important features 5/19
6. Hating on the jet colour map 6/19
7. Hating on the jet colour map 7/19
8. Hating on the jet colour map 8/19
9. Hating on the jet colour map 9/19
10. Hating on the jet colour map 10/19
11. What have we learned? • Jet is not a great colourmap (or is it?) • Two types of information one can glean from a colourmap1 • ‘Value’ or ‘metric’ information • ‘Form’ or ‘structure’ information • Jet is not bad for value information (but not everywhere) • Jet is awful for form information • Jet is not alone—but it is very commonly used • How to pick a good colour map? It depends! 1C. Ware, Color sequences for univariate maps: theory, experiments, and principles, IEEE Computer Graphics and Appliations, 1998. 11/19
12. What is good for form/structure information? • Colour has 3 dimensions: hue, saturation, and luminance • Saturation-varying colourmaps are good for low-frequency data • Luminance-varying colourmaps are good for high-frequency data • The human brain is very bad at interpolating hue2 • Perceptually-based colourmaps • Equal steps in data are perceived as equal steps in the colour space 2Conclusion from psychophysical experiments by S. S. Stevens (formerly at Harvard) 12/19
13. Perceptually-based colourmaps 13/19
14. Perceptually-based colourmaps 14/19
15. Perceptually-based colourmaps 15/19
16. Perceptually-based colourmaps 16/19
17. Perceptually-based colourmaps 17/19
18. Perceptually-based colourmaps 18/19
19. Thank you. Pssst. Jet is the default colourmap in matplotlib. Anybody want to fix it? Submit a PR! Link to slides: https://github.com/dmcdougall/scipy14-colormaps 19/19