Re: A color tutorial from Tom Jewett

Hello David, Bruce, Wayne, Jonathan, John, everyone,

Some of you may know me from Issue #695, the GitHub issue I started two months ago on this subject. (Yes, I admit I’m the one that hit this hornet’s nest with a baseball bat LOL). I just posted a message to Gregg, but I wanted to address some of the other comments here as well.

First, I am now a member of W3C and Low Vision Task Force as an invited expert, I look forward to working more directly with all of you. Responses to your recent posts regarding contrast are all inline in this message, below.

For those of you that are interested, the first round of research was summarized (including experimental results) in that thread. LINK: https://github.com/w3c/wcag/issues/695 <https://github.com/w3c/wcag/issues/695>.  also tangentially involved in issue 665.

And I have “live/semi-live” experiments going on here: https://www.myndex.com/WEB/Perception <https://www.myndex.com/WEB/Perception>

Also on my ResearchGate account, though not so busy there at the moment.

PLEASE FEEL FREE TO COMMENT ON THE EXPERIMENTAL TESTS, send those comments to me directly if you like.

I’m going to answer everyone cc'ed in this single email to group the comments together:
________________________________________________________________
________________________________________________________________


> On Jun 7, 2019, at 6:57 AM, David MacDonald <david100@sympatico.ca <mailto:david100@sympatico.ca>> wrote:
> agreed.... 
> Is anyone aware of serious research being conducted in this area, including trials with significant variations of users that might result in proposed amendment to the algorithm? Currently, I'm unaware of any serious initiatives.

I am doing serious research on this, and have been for months. The most recent publicly displayed experiments CE14 and CE15 have some very encouraging results, but I am still finalizing a few minor details.

At the same time I am developing a study (more of a study set actually).
 
Part A: Internal Evaluations. This is ongoing right now with the experiments on the Perception page <https://www.myndex.com/WEB/Perception>. This is a more informal review and discussion. Feel free to get involved. The current experiments right now are CE14 and CE15, though soon to be replaced — just look at the bottom one or two on the list, and then please email comments direct to me. NOTE: At the moment the pages are NOT responsive and will break on a phone (this is because I am using REM font sizes). I will be making a responsive version soon.

Part B: Mass public study. This is the one I am developing at the moment, a web-based public perception study. This will be unsupervised. While it will lack a controlled, it has some advantages. First, sample size. Ideally it this study will collect thousands of subjects, and with a large enough sample the uncontrolled element should smooth out any far outliers. Second, people will be using their computer/device in their normal environment, under their typical conditions.

Part C: Individual subjects. In this part, I’ll either set up a perception lab here, or take my laptop to locations to conduct trials in person with individuals. I am in the Los Angeles area, and I have not reached out to any vision impaired potential test subjects yet, so volunteers will be sought at some point in the near future. I intend the Part A study to take about 20 to 30 minutes per subject.

A
________________________________________________________________
________________________________________________________________

> On Fri, Jun 7, 2019 at 7:19 AM Bruce Bailey <Bailey@access-board.gov <mailto:Bailey@access-board.gov>> wrote:
> Yes, absolutely!  I see that I did write “main purpose” below (and that was a mistake), but what I really meant to be emphasizing is the genesis of the formula came from addressing color blindness, where specific difficulties and remediation strategies were well understood.  Then it turns out that paying attention to luminosity contrast ratios helps will all kinds of low vision issues.  Certainly many people do not find the contrast sufficient as the ratio approaches 4.5:1.  But in terms of the algorithm allowing for false positives, and in the absence of clinical trials like the Lighthouse did back in the day, I still think it is useful reassurance to test the edge cases against other maths that we have (e.g, Protanopia, and Deuteranopia).
> 
> 

Luminance contrast (or a perceptual variant like L*) is key for many users. Some studies show that color contrast can interfere with impaired vision, thought the mechanisms are not well understood. From PSYCHOPHYSICS OF READING: XI

Low-vision reading rates are lower for color contrast than for luminance contrast. This was true at maximum contrast but increasingly so at lower contrasts. Relative to normal subjects, low-vision reading is hampered, not enhanced, by color contrast.
Why should color contrast be more deleterious to low-vision reading than to normal reading? Some of our low-vision subjects complained of glare when reading the color-contrast text. 

In our past work, we have shown that subjects with cloudy media read white-on-black text faster than conventional black-on-white text because of the extra light scattered from the page in the latter case (Legge et al., 1985b; Legge et al., 1986). A similar explanation might account for depressed reading of equiluminant text. In this case, light can be scattered from both letters and background to dilute retinal-image contrast. It was previously shown that glare effects in low vision are not restricted to subjects with cloudy media but are found widely in subjects with clear media and central-field loss (Rubin et al., 1986). Whatever the neural explanation in such cases, glare may play a role in explaining depressed reading with color contrast.

SUMMARY
For normally sighted subjects, reading rates for high color contrast are as fast as those for high luminance contrast, more than 300 words/min.

I’ve been planning trials & developing the web app this week, among other things. 

As for contrast: perceived contrast is so very variable! I have studies where increasing beyond 5:1 had no measurable effect, and other studies that normally sighted subject read fastest with a 10:1 contrast. Ambient light, font size and weight, padding around text all have such a strong effect, and yet so many studies do not define these specifications.

As for CDV Protan/Deuter — red and green cone responses overlap substantially. Protan (no/weak red) has the problem that there are no other receptors at longer wavelengths, so pure reds get dark or black. But otherwise the green cone responds to much of the red wavelength. The missing Green of the Deuter types though is surrounded by red, blue, and rods. While the green/cyan will be a little darker, the red cone will respond to much of the green stimulus due to the large overlap.

Green is 71% of luminance in normal vision, and red is 21%. If you are missing red cones, the green cone  will respond to much of the red light (down to about 620nm), but you’ll likely see pure red as black at lengths longer than 650nm, so site using a red and black theme might be a problem. 

The thing is the current math, and most of the replacements, are color agnostic. All the contrast math takes place after the colors have been combined into Y luminance, so there is no consideration for color at that point. Losing green in deuter, the red cone takes up most of that light/luminance.

My point being aside from red/black for Protan-types, the four most major CVDs are well handled by “color free' luminance. 

A
________________________________________________________________
________________________________________________________________

> On Jun 6, 2019, at 3:46 PM, Jonathan Avila <jon.avila@levelaccess.com <mailto:jon.avila@levelaccess.com>> wrote:
> 
> t the main point of the contrast requirement is to address colorblindness issues – while that is an important issue as having sufficient contrast provides a non-color way of distinguishing content – having sufficient contrast is extremely important for users with low vision. Sufficient contrast levels are needed for a number of vision conditions and contrast sensitivity tends to decrease with age.  Not all people with low vision need nor want high contrast – but a sufficient level is needed and is enough for many to access content.

Yes, luminance contrast (or similar perceptual lightness L*) is important for everyone. While normal vision folks can read well with only color contrast, there are a lot of color contrasts with minimal luminance contrast that can be a problem for *everyone*.

Here’s an experiment I did a few months ago : https://www.myndex.com/web/huecontrastchart <https://www.myndex.com/web/huecontrastchart>  This is a color contrast chart, with much of the luminance contrast removed. Scroll it up and down or side to side quickly: notice the black lines that appear perpendicular to the motion on so many tiles? About 35% of these colors are in opposition, and edge detection is on high alert. The only oppositional color that *isn’t* that bad is yellow/blue, but I believe that is because that color pair ALSO has the highest luminance contrast of any pair on the chart.

THE POINT: using colors in opposition (especially green/red) without ALSO using a strong luminance contrast pushes the oppositional aspects of human vision — for web content this can create a number of artifacts, and especially with hand held devices that are constantly moving. Even for normally sighted, luminance contrast is a good design “best practice”.

CONTRAST SENSITIVITY:

Contrast sensitivity relates to threshold contrast, which is 1% for normally sighted (a Pelli-Robson score of 2)  3% for “becoming impaired”  (a Pelli-Robson score of 1.5) and perhaps 10% for a profound impairment  (a Pelli-Robson score of 1).

From: https://www.ncbi.nlm.nih.gov/books/NBK207559/#ddd00103 <https://www.ncbi.nlm.nih.gov/books/NBK207559/#ddd00103>

The Pelli-Robson score represents the logarithm of the subject's contrast sensitivity. Thus a score of 2, indicating a contrast sensitivity of 100 percent, means that the lowest contrast letters the observer can read correctly have a contrast of 1 percent (i.e., 1/100).

Whittaker and Lovie-Kitchin (1993) surveyed the literature on the effects contrast, on reading speed. They defined the “contrast reserve” as the ratio of print contrast to threshold contrast. From their survey of the published data on low and normal reading rates versus text contrast, they concluded that the contrast reserve had to be at least 10:1 for reading at a low normal speed of 174 wpm; a 4:1 reserve to read at 88 wpm, and a 3:1 reserve for “spot reading,” i.e., 44 wpm. 

For newsprint with a contrast of 70 percent, then the reader's contrast threshold would have to be lower than 7 percent to achieve the desired 10:1 reserve.

All that said, NASA and the FAA advise against a contrast MORE THAN 15:1, such high contrasts can cause scatter, glare, and chromatic aberration. They also advise a standard contrast of 7:1 for most situations, with 3:1 a minimum for 20/40 and better, and 2:1 allowable in some operating circumstances.

A
________________________________________________________________
________________________________________________________________

> On Jun 7, 2019, at 10:43 AM, Wayne Dick <wayneedick@gmail.com <mailto:wayneedick@gmail.com>> wrote:
> 
> I think we can all relax. We know contrast sensitivity is an almost universal issue. Even people with photophobia will frequently opt in favor of higher contrast. The only question is this. Is the formula optimal for the technologies today?

The formula wasn’t optimal for technologies of yesterday, much less today. Hard even to call it a formula, it’s just a simple ratio. 

CURRENT:
  C = (Lhi + 0.05) / (Llo + 0.05)  


CANDIDATE 4 - MW:
  C = (Lhi - Llo) / (Lhi + 0.1)   

I have some more promising candidates, and more functional ones - but I want to tweak them before bringing them forward.

> We cannot do a change today. It will take study and comparison. Maybe we should look at the past and current studies and re-run them with a few tests. Something like this cannot be introduced until Silver. 

Yes, which is why I have my experiment pages up, and am working on concepts and testing/study protocols at the moment.


> The research for this will take some time, but it is worth doing.


Perhaps. I have multiple working candidate solutions, and I done some testing in adverse (i.e. bright outdoors) environments. There are other things planned/in pipeline as I mention in some other replies herein. I have a somewhat ambitions study/testing/data gathering concept that is probably going to be the longest part of the process.

A
________________________________________________________________
________________________________________________________________

> On Jun 7, 2019, at 10:00 AM, John Foliot <john.foliot@deque.com <mailto:john.foliot@deque.com>> wrote:
> 
> A few thoughts,
> Wayne wrote: "Has the IEEE taken any recent steps to clarify how to compute cross platform computations of luminance?" 
> 
> I'd expand that to a) "has *anyone* taken any recent steps..." (sourcing is less important than good data), and / but b) part of the issue with screen luminance is that there is no "standard" for individual screens, and so any measurement there would be tied to the tester's screen, which may or may not be appropriate for all screens.

Well, there IS a standard, it’s:
  
IEC-4WD <>
IEC/4WD 61966-2-1: Colour Measurement and Management in Multimedia Systems and Equipment - Part 2.1: Default Colour Space - sRGB. <>
It’s just that few people set their display as dark as 80 cd/m^2 LOL. But otherwise, that is the standard for the web and most computer systems, at least for the time being. 


> Even if it is a pure-play "mathematical" algorithm, it's nearly impossible to account for end-user configurations, which will have a direct impact on "measurement versus practical usefulness”.

While you can’t account for “exact” user settings, settings will be in a fairly defined range. 


> (Walk into any Best Buy, Target or Walmart and go to the "TV Department" - multiple screens with the same broadcast, yet color and luminance differences are usually very obvious),

Yea, TV stores usually have the controls all cracked to make every set as bright and contrasty as possible. BLECH.


> and so meeting a specific luminance value may or may not aid individual users.

Not sure what you mean here, but hitting a specific luminance value is not as important as the relative screen luminance to the ambient light. 


> I often recount the scenario of taking your cell phone outdoors at lunch time and trying to access your (test) site iunder those conditions. There, the influence of the environment (bight sunshine) has an impact on luminance measurement and perceivability (i.e. it is a conditional measurement). I'm not sure how we can adjust our algorithm to address that variable.

Experiments CE11 through CE15 (on the Perception link) have been tested in real world, laptop outdoors in mid-day sun conditions. 


A
________________________________________________________________
________________________________________________________________

>  From: David MacDonald <david100@sympatico.ca <mailto:david100@sympatico.ca>> 
> 
> Sent: Thursday, June 6, 2019 3:13 PM
> 
> 
> I think some people with low vision are disproportionately affected by perceived low contrast. 
> 
> 
>  Having said that, I don't know what would need to be done to the algorithm to improve it, and without, as Gregg said, significant input by knowledgeable researchers, like Lighthouse did back in the day, who are sensitive to the great amount of research put into our current algorithm, with a proposal and a demonstration of the benefits of the improvements, I wouldn't revisit the algorithm. 
> 


Most of what you are asking I discuss in the many, in-depth, posts in issue 695.

https://github.com/w3c/wcag/issues/695 <https://github.com/w3c/wcag/issues/695>.

> There are, however, a few color combinations that seem a little weird to me. Usually, a threshold contrast (4.5:1) it is when one is black. I think there has been a little buzz around that in the public.

I’ve read a lot of disparaging comments from designers regarding the methods in general. I hope to see that attitude change with a better implementation that designers are more willing to use and work with. If they don’t like it and are not willing to use it, it does no good.

> I'd like to see a study identifying these types of combinations that seem harder to see than some failing contrasts, with suggestions on adjustments we can adjust the algorithm.


Again, detailed at length in issue #695 and also on the perception link in some of the live experiments. 

A


—End