apca-introduction

commit

f36acc6c64f00132e9fa2c96bbddf7e361d89c5a

parent

13c1be744cbdd4f9079afaa2a2911a477460b63d

Author

Tobias Bengfort <tobias.bengfort@posteo.de>

Date

2023-09-14 14:41

tweak analysis

Diffstat

1 files changed, 9 insertions, 15 deletions

diff --git a/analysis.md b/analysis.md

@@ -339,8 +339,8 @@ The most complete (but also most complex) color appearance model currently
  339   339 available is [CIECAM02](https://en.wikipedia.org/wiki/CIECAM02). It uses
  340   340 exponents between 0.31 and 0.72. Given that model, WCAG 2.x is on the lower
  341   341 (darker) end of possible exponents, while APCA goes to the other (lighter)
  342    -1 extreme. Choosing the right parameters depends on which lighting conditions we
  343    -1 want to model, and that is ultimately a policy decision.
   -1   342 extreme. This is consistent with the observation that APCA reports lower
   -1   343 contrast for darker colors.
  344   344 
  345   345 It seems reasonable to use a lower exponent for light-on-dark color pairs.
  346   346 First, because the background color itself is often a significant part of the
@@ -541,8 +541,8 @@ algorithm in many key aspects:
  541   541 
  542   542 -   It uses a different luminance calculation that deviates from the standards
  543   543     but is supposed to be closer to real world usage.
  544    -1 -   It uses the more accurate Stevens model and significantly different
  545    -1     parameters for converting luminance to perceptual lightness.
   -1   544 -   It uses the more accurate Stevens model and assumes different lighting
   -1   545     conditions for converting luminance to perceptual lightness.
  546   546 -   It adds an additional step where different exponents are applied to
  547   547     foreground and background.
  548   548 -   It uses different scaling. Crucially, this scaling is based on a difference
@@ -550,19 +550,13 @@ algorithm in many key aspects:
  550   550 -   It uses a more sophisticated link between spatial frequency and minimum
  551   551     color contrast that might allow for more nuanced thresholds.
  552   552 
  553    -1 The new contrast formula agrees with WCAG 2.x for 83.9% of randomly picked
  554    -1 color pairs. That number rises to 92.5% for a modified WCAG 2.x formula with a
  555    -1 flare value of 0.4. As far as I understand, this is not a realistic value for
  556    -1 flare. So the physical interpretation might be incorrect. This would however
  557    -1 explain why APCA reports lower contrast for darker colors.
  558    -1 
  559   553 So far I like many of the ideas of APCA, but I am not convinced that they are a
  560   554 significant enough improvement to justify breaking backwards compatibility. I
  561    -1 am also concerned by the [lack of publicly available evidence]. Then again, the
  562    -1 new algorithm cannot really be evaluated without first making some policy
  563    -1 decisions, e.g. which viewing conditions we are aiming for. I hope this
  564    -1 analysis can support the community in figuring out what questions need to be
  565    -1 answered.
   -1   555 am also concerned by the [lack of publicly available evidence].
   -1   556 
   -1   557 Much of the difference between APCA and WCAG 2 comes down to a different choice
   -1   558 of parameters, and that is ultimately a policy decision. I hope this analysis
   -1   559 can support the community in figuring out what questions need to be answered.
  566   560 
  567   561 [Web Content Accessibility Guidelines]: https://www.w3.org/TR/WCAG21/
  568   562 [sRGB color space]: https://en.wikipedia.org/wiki/SRGB