Re: coverage to alpha

[Eric Muller]

> Coverage == opacity (aka alpha). It is more than plausible, it
> is justifiable.

And the justification is?

---

Consider a vertical stem, 3px wide. When the edges are aligned to the 
pixel grid, we have coverages [... 0 1 1 1 0 0...] along the x axis and 
that creates a certain perception of thickness. If we shift that stem 
progressively in x, we get coverages [... 0 c 1 1 (1-c) 0 ...]. Our goal 
is certainly to create the same perception of a stem, with the same 3px 
thickness. That is going to involve some kind of 
processing/interpretation by the visual system since the stimulus is 
different.

So 1: we cannot ignore our visual system when we implement text display 
systems.

Looking at White's illusion, it is clear that the same gray, depending 
on its context, is creating different color perceptions. (There are many 
other illusions, I just picked that one because its manifestation seems 
to be universal, it is rather brutal, and is a scale that does not 
involve pixels).

The Westheimer quote is also pointing out that our visual system is not 
trivial, and that time (learning) may also come into play.

It is also well known that using the same alpha map, white text on black 
background does not create the same perception as black text on white 
background.

So 2: our visual system works in counter intuitive ways.

> I do not know why you are concerned, or how, or why you want to
> compensate for this using coverage or alpha.

If we consider the typical rendering process as a whole:

- from outlines positioned and scaled on the pixel grid to a coverage map
- from a coverage map to a linear alpha map
- from a linear alpha map to pixels

it seems to me that by construction, i.e. by the definitions of coverage 
and linear alpha, the first and last steps are entirely mechanical.

So 3: in that process, the only place we can inject something to account 
for our visual system is when going from coverage to alpha.

> FreeType prefers to stay out of it. We sell *coverage*. Take it or
> leave it. Once bought, you can do whatever you want with it in the
> context of your perception experiments :)

Sorry I engaged the wrong discussion list.

Eric.


On 6/8/21 1:27 PM, Alexei Podtelezhnikov wrote:
> CAUTION: This email originated from outside of the organization. Do not click 
> links or open attachments unless you can confirm the sender and know the 
> content is safe.
>
>
>
> On Tue, Jun 8, 2021 at 11:14 AM Eric Muller <emuller@amazon.com> wrote:
> > On the one hand, we have a coverage map, i.e. for each pixel how much of it is 
> > covered by the outline.
> >
> > On the other hand, we are blending a foreground (text) color with a background 
> > color. So we need some alpha, and use it for a linear blending, in a linear 
> > colorspace.
> >
> > What is not clear to me is how to go from coverage to alpha. Identity is 
> > plausible, but there are also reasons to believe it is not:
>
> Coverage is a pixel proportion that draws the outline foreground
> color: Coverage == opacity (aka alpha). It is more than plausible, it
> is justifiable. It certainly helps to use gamma correction too so that
> blending is linear.
>
>
> > - things like White's illusion show that the perception works in strange ways
>
> I do not know why you are concerned, or how, or why you want to
> compensate for this using coverage or alpha.
>
>
> > - "One usually begins by assuming that nothing is known about the object world and 
> > then the diffraction limit outlines the range of object details that an image transfer 
> > allows to be gained and, by exclusion, those that it leaves undetermined. On the other 
> > hand, it might be known ahead of time that the ensemble of possible objects is 
> > restricted. Then distinctions can be made by concentrating on the expected differences 
> > and disregarding image aspects that might have arisen from sources known beforehand to be 
> > absent." (Optical superresolution and visual hyperacuity, Westheimer), which can 
> > explain how readers can perceive gray pixels differently (i.e. expecting black and white, 
> > and therefore perceiving gray as width)
> >
> > - may be the alpha should also depend on the foreground/background color
>
> FreeType prefers to stay out of it. We sell *coverage*. Take it or
> leave it. Once bought, you can do whatever you want with it in the
> context of your perception experiments :)
>
>
> > - may be the alpha should also depend on the ppem.
>
> Again, we sell coverage. It looks great as alpha with gamma correction.