I am not colorblind, but work with data visualization in Matlab a lot. This hilarious SciPy2015 talk[1] convinced me to switch to the Parula color scheme. There's no way I'm going back to Jet.
Yes, Viridis is quantifiably better than Parula, as the talk shows, but Parula comes with the standard Matlab, and the jump from Jet to Parula was of such a tremendous magnitude that I kind of fell in love with the Parula color scheme right away. Jump from Parula to Viridis would be less noticeable, I guess.
The talk opened my eyes to the fact that there are much better ways to use color for information presentation. (Sorry for the late reply)
This talk was awesome. Best explanation i've ever heard for why color spaces have the odd shape they do. The 3d models of perceivable color spaces for different colorblindnesses was also really insightful.
I've been using Color Oracle [1] to simulate colour-blindness as part of my standard developer workflow recently. For such a common disability, it's shocking how little work most people put into ensuring accessibility for colour-blind people.
If you have an Android phone, you can also go into the developer settings, turn on a colour blindness simulation, launch the camera, and then point it at anything (in the real world, or on a monitor) and see an idea of what it might be like to a colour blind person.
I am always pleased to see annoucements and discussions on colour blindness.
Forgive me for sharing again my writeup on this syndrome from my own experience, if you don't yourself have any such deficiencies you might find it interesting!
Ha! I enjoyed this read. I am also colorblind. I do think one of the more interesting parts of being colorblind is the fact that I do not rely on color for ANYTHING. For example - Stoplight? What position is lit is my first thought. I think this train of thought has had some unintended benefits in terms of my ability to process things analytically
Thanks! (If you ever see them in a super pale pink shirt, check if they believe it's white and may've been ruined in the laundry. Been there, done that)
About 8% of all males are color-blind. If you are a scientist, and want to publish your work, there's a one in four chance that at least one of your three reviewers is color-blind.
Rainbow color schemes are ridiculously bad. One study showed that medical professionals who look at rainbow-themed plots every day, immediately made fewer errors if presented with a simple grayscale plot instead, despite not having any experience with the grayscale color scheme.
Add to that that human perception is much better at seeing high-frequency lightness contrast than color contrast. Color contrasts are better-suited for categorical data.
This reminds me of CubeHelix [0], which aims to create scales which degrade gracefully to grayscale, which also minimizes potential color-blindness issues.
With just two colours it's hard to see the intermediate values because our eyes tend to blur together colours; is that an intermediate value or sharp edge that I'm seeing fuzzily due to poor focus? With an intermediate tone, such issues are pushed down to lower resolutions - with yellow-red-blue I know for sure that if it's red it's an intermediate, but is the orange area a mix of lots of small yellow and red domains or is it a field of intermediate domains?
Rinse and repeat.
The article seems to say "this scale is better" whilst also describing how other scales can be better. The evidence they present says to me "choose scales according to the data, the desired use, and the observer", of you ship the data used to cover a visualisation along with that visualisation then people can use their own scale altered to the purposes (and disabilities) they have.
Any scale that only uses hue is terrible, no matter how many intermediate steps. Scales that only accidentally utilize luminance (due to human vision being most sensitive to green, then red, and by far the least sensitive to blue) are only marginally better. Indeed, the rainbow scale is atrocious also due to the perceptually most luminous color being somewhere in the middle of the range, so people with red-green deficiency cannot even easily use luminance cues to make sense of the data. Not even mentioning the fact that human vision is simply much better at distinguishing luminance differences than hue differences.
I just skimmed the paper [1], and their conclusion starts:
> We identified one colormap in particular to be optimal for viewing by those with or without CVD, which
we name cividis (Figs 4 and 5), generated by optimizing the viridis colormap and selecting the J'
linearization that maximizes the range of J'. We chose this map due to its wide range of colors, resulting
from a wide range of J0 values while still changing b0 significantly, and overall sharpness when overlaid
onto complex images.
I'm disappointed that they (in my opinion) managed to make a worse color scale. If you look at the CVD-Jet color bar you can see blocks that appears quite similar separated by too sharp transitions. This is a common problem in e.g. rainbow scales. They highlight them selves that yellow appears as highligt, yet the highest values get translated to black. How is black supposed to be interpreted as more luminescent/intense than yellow? It happens to be a color scale that makes the cells look good, but to me that is happenstance.
CVD-Jet is not the new color scale, it's a simulation of what Jet looks like for certain colorblind users.
To my eyes, cividis (which is the new scale) is good by the numbers, and there's a strong argument it's more robust for colorblind users, but viridis looks better.
> The jet colormap is associated with an astrophysical fluid jet simulation from the National Center for Supercomputer Applications. See the "Examples" section
Apart from the comment on line 1, that file contains 256 lines with a colour. I suppose to get a finer scale, you can interpolate with any reasonable function you like since the differences are so small.
Semi-related tid-bit: You'd be shocked how many theatrical lighting designers over the age of 45 are color blind. Never used to be a workplace issue since there was only a limited number of gels to place over lights and even then they were all numbered / catalog.
With LEDs being able to project (nearly) any color the old-guard are facing some challenges.
They claim that the new scale gives a continuum from dark to light, but if you look at the example it goes dark at both ends with the brightest colors in the exact middle. Why the discrepancy?
I don't think it does. The picture here: https://static.scientificamerican.com/sciam/assets/Image/UPj... is CVD-Jet, not their new scale. The blue-and-yellow rightmost picture is described as "as a person with red-green color blindness sees the rainbow image [in the center]". This scale does put the brightest colors in the center.
Correct, one of the links in the article shows a pretty clear comparison between rainbow scale, how that looks to someone with color deficiency (with the dark portions on the ends), and their proposed scale.
The image at the top of the page loads fine. But the rest an so encumbered with Javascript to load you literally have to reload the page at least 4 times (to load the JS domain, that loads the JS domain, that loads the JS domain) to see them.
[1] https://www.youtube.com/watch?v=xAoljeRJ3lU