W3C home > Mailing lists > Public > w3c-wai-gl@w3.org > April to June 2007

Contrast and colour deficiency

From: Sean Hayes <Sean.Hayes@microsoft.com>
Date: Thu, 14 Jun 2007 13:39:00 +0100
To: WCAG <w3c-wai-gl@w3.org>
Message-ID: <7261AC2A5F73904CA41773C8F00813FF31D79FA4@EA-EXMSG-C309.europe.corp.microsoft.com>
One of my assigned items this week involved the issue of colour combinations passing at level A, which would be problematic for users with atypical colour sensitivity. Here is more background to the item surveyed this week.

I have attempted to study the issue a little to see if it is a real problem, or just a theoretical one, and have built a tool which (if I've understood everything correctly, which is a big if) allows me to determine the contrast for users with atypical colour sensitivity.

I've also been using the tool provided by Gez Lemon (although my results differ slightly from his, but I'm working from 0-255 sRGB, whereas he is working off hex numbers, so it might be a rounding error, however Gez if you are out there can you contact me to explain the math you are using on your tool, so I can figure out if I'm doing something wrong)

 I think at marginal situation there is indeed an issue; for example if I choose:
Foreground        0              0              0
Background        207         0              47

I get a computed ratio of 3.69 (Gez seems to get 4.1),  which is acceptable under our provision for 18pt text.

whereas for a protanope the colours perceived would be:
Foreground        0              0              0
Background        83.12     72.46     49.35

(Using the simulation method devised by J.D. Mollon et al)

For these colours the computed ratio comes out as 2.35 which would fail the contrast provision.

So assuming that a protanope would see a different contrast in this manner, we seem to have a discrepancy.

Sometimes however this works in reverse, for example red 159,0,0 text on green 0,255,0 background has a contrast ratio of 6.12 (5.65:1 according to Gez) thus passing our contrast provision at AA. But for a protanope, by my simulation, the perceived contrast is over 10. However for a deutranope it drops to 4.9, thus failing for small text.

For greys:
Fore     255      255      255
Back    110      110      110

Normal Contrast         5.098907547
Protanope Contrast     4.645289213
Deutranope Contrast   4.619764382

So it seems we do indeed have a problem.

My initial thought was that this should be covered by 1.4.1 which deals with colour, but this does not help in this instance because the information is not being conveyed by the colour of the text, but the text itself.
The contrast section is not catching this properly. So I believe we need to consider a new provision which deals with hue difference as well as the existing contrast difference.

One option is to increase the basic minimum luminosity contrast, e.g. to 7 or more; which should then raise the bar sufficiently that the reduced contrast in atypical colour reception would be raised above the existing minimum bar. I think this approach is problematic for two reasons, one it starts to significantly impact on site design, for example when one needs to use a background and an enabled and disabled colour. Secondly without more investigation I'm not sure whether the idea is in fact sound in the first place, as some colour combinations may have significantly worse deterioration of contrast than others and is this deterioration is probably condition specific.

We could define a second specific luminosity contrast for some hues, which could be more readily achieved at level A. This approach could work, but I think this gets very complicated when we try to deal with all of the forms of colour deficiency.

We could add a whole new provision, such as (this is slightly modified from the survey question):
"Text (and images of text) do not cause [ambiguous hue difference], except if the text is pure decoration."
And define [ambiguous hue difference] as : A pair of adjacent colors, where the perceived colors for dichromat, or anomalous trichromat vision have a [luminosity contrast ratio] less than n:1 (where n= 5 at AA, and 7 at AAA).

I think this has the most promise, the clause in the normative section basically says 'consider people with colour deficiencies', which should be enough to get authors thinking along the right lines, and there are plenty of resources on the internet to help. For the definition itself, we probably need to contact some experts in colour deficient vision in order to get the exact wording correct, but we have fudged this a bit in other areas so maybe we can here too .

But perhaps the simplest approach is to fold the definition into the existing contrast definition, adding a note that this is not considering monochromat vision, since (from what I can gather from the literature), a) it is quite rare, and b) it is generally accompanied by significant other visual loss which implies assistive technology is a more appropriate solution.

contrast ratio

Is the minimum ratio, determined for a standard observer, for dichromat vision, and for anomalous trichromat vision using :

(max(L1,L2) + 0.05) / (min(L1,L2) + 0.05), where
·         L1 is the relative luminance<http://www.w3.org/WAI/GL/WCAG20/#relativeluminancedef> of the perceived foreground color,
and
·         L2 is the relative luminance<http://www.w3.org/WAI/GL/WCAG20/#relativeluminancedef> of the perceived background color.

          [reference Computerized simulation of color appearance for dichromats: Brettel,  VieŽnot and Mollon].
Received on Thursday, 14 June 2007 12:40:00 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:47:50 GMT