Understanding LMS-based Color Blindness Simulations | DaltonLens

[utterances-bot]

Understanding LMS-based Color Blindness Simulations | DaltonLens

Diving deeper in the theory of Viénot, Brettel & Mollon’s method to understand what’s behind the scene.

https://daltonlens.org/understanding-cvd-simulation/

[warren-gavin]

I didn't think I was going to spend my Sunday reading three of the most complete explanations of colourblindness and its simulation. I tried reading the 1997 Brettel paper a few years ago, but I didn't have the domain knowledge to understand any of it.

This blog post has been a revelation, it's a great explanation of the subject matter. I never understood how the colour spaces could be reduced to a 2D plane, but now it makes perfect sense.

I love the concept of confusion lines. The generated 3D plots helped me realise I've protanopia, where all my life I thought I had deuteranomaly.

Thanks for a great resource Nicolas!

[nburrus]

I'm sorry for your Sunday, but glad you found the articles useful :) Thanks for the feedback, much appreciated!

[page200]

What result do the simulations try to produce:

• Is the simulated image intended to be viewed by a non-colorblind person, such that the amounts of perceived differences between all pairs of colors in the image are similar to those perceived in the original image by a colorblind person?
• Or does the simulation try to make the original and the simulated image look similar to each other when viewed by a colorblind person?
• Or a compromise between these two goals? Or do both goals yield the same results?

[nburrus]

Short answer: both :)

Long answer: the original and the simulated image are expected to look similar to a colorblind person (if he's a full dichromat). However there are many possible simulated images that would look similar to the original image for a colorblind person, so the generated one is the result of a number of assumptions, some of them based on observations in unilateral dichromats (people with one normal eye and one colorblind eye).

For a normal observer, the simulated image allows to see how much color information is lost. A colorblind observer indeed cannot differentiate more colors in the original image than a normal observer on the simulated image. Now really knowing whether a colorblind person "perceives" the original image the same way a normal observer "perceives" the simulated image, no one can really say.

[page200]

Short answer: both :)

In the sense that one result achieves the optimum for both goals simultaneously, or that the result is a compromise between how well each goal is satisfied?

Long answer: the original and the simulated image are expected to look similar to a colorblind person (if he's a full dichromat).

I wonder about the "full dichromat" part. So a person with deuteranomaly (not dichromat) wouldn't perceive the deuteranomaly simulation as similar to the original image? So the goal number 2 I listed earlier is targeted only in simulations of full dichromacy, whereas goal 1 is targeted in all simulations?

However there are many possible simulated images that would look similar to the original image for a colorblind person, so the generated one is the result of a number of assumptions, some of them based on observations in unilateral dichromats (people with one normal eye and one colorblind eye).

• Re "some": What assumptions aren't based on observations in unilateral dichromats?
• Regarding the assumptions that are based on observations in unilateral dichromates, do you mean that simulations of dichromacy try to not change colors that unilateral dichromats perceive similarly with both eyes?

In the article, you write "the similar colors for the two eyes are observed along the 475nm (blue-greenish) and 575nm (yellow) axes for protans and deutans, and along the 485nm (cyan) and 660nm (red) axes for tritans". Could you please visualize these axes?

For a normal observer, the simulated image allows to see how much color information is lost.

How is that defined? That a non-colorblind observer perceives simulate_cvd(color1) and simulate_cvd(color2) as similar, iff a colorblind person perceives color1 and color2 as similar?

A colorblind observer indeed cannot differentiate more colors in the original image than a normal observer on the simulated image.

"Not more" sounds like the "≤" sign. Can the colorblind observer differentiate fewer colors? Or approximately the same?

Now really knowing whether a colorblind person "perceives" the original image the same way a normal observer "perceives" the simulated image, no one can really say.

Indeed, qualia and so on. :)

Thank you!

[nburrus]

In the sense that one result achieves the optimum for both goals simultaneously, or that the result is a compromise between how well each goal is satisfied?

Yes both goals are achieved at the same time. I don't see a tradeoff that could improve one objective while hurting the other.

Re "some": What assumptions aren't based on observations in unilateral dichromats?

An example is that the projection surfaces are linear planes and not curves or something fancier. But there are many other sources of approximation. If you want to go deeper the discussion section in the Brettel et al. paper mentions some of them.

This article Broackes,2010 discusses that planes are probably not a very good model and more generally gives a critical view of the unilateral dichromacy observations everyone refers to.

Regarding the assumptions that are based on observations in unilateral dichromates, do you mean that simulations of dichromacy try to not change colors that unilateral dichromats perceive similarly with both eyes

Yes exactly.

"the similar colors for the two eyes are observed along the 475nm (blue-greenish) and 575nm (yellow) axes for protans and deutans, and along the 485nm (cyan) and 660nm (red) axes for tritans". Could you please visualize these axes?

They are kind of shown in the Projection planes in LMS section. These points are in the plots, and the planes go through them.

How is that defined? That a non-colorblind observer perceives simulate_cvd(color1) and simulate_cvd(color2) as similar, iff a colorblind person perceives color1 and color2 as similar?

What I meant is that if the colorblind observer cannot differentiate two colors in the original image, then the normal observer won't be able to differentiate them either in the simulated image. So this allows the normal observer to see what color contrasts are lost for a colorblind observer. Which is the main practical utility of these simulations. If a normal observer can't differentiate e.g. the color of the different players of a game in the simulated images, it means a colorblind person will have a hard time playing. And the reverse should be true too, if a normal observer can play with the simulated images, then a colorblind person will be fine too.

"Not more" sounds like the "≤" sign. Can the colorblind observer differentiate fewer colors? Or approximately the same?

A full dichromat observer won't differentiate less colors, but the same amount, so it's really an = sign.

[Waitsnake]

Thanks a lot for explaining all those different color spaces and how the are defined and how to convert between them! Also for introducing the concept of the 'confusion lines' to me in 3D since in the past I just saw them drawn on some 2D planes where I could not understand how this should confuse me as a protan. On your 3D graphs it is easily visible to me that I have only problems with differentiating to colors along the lines of the the 'protan confusion lines' and but not on the other confusion lines for deutans and tritans. Those lines help also to see how good a daltonize correction works to compensate this issue (I know it can be only a compensation no correction).