Lab CAT Logic

[BradFunkhouser]

Is CIE 1976 Lab just a math transform to and from XYZ that yields a more perceptually uniform space relative to a specific white point, or does that transform also provide a chromatic adaptation function which makes Lab a form of appearance model between different white points?  I've seen it presented both ways and it's confusing.  Here's my current understanding.  Please correct me if I'm wrong.

Say I have two printer profiles with different paper white points.  They both assume a D50 illuminant is being used to view the prints.  A Lab color inside the first printer/paper space will have it's own specific XYZ value.  If I use those same Lab values in the other printer/paper space, that color will now have different XYZ values because the white point is different.  If I get fully adapted to the first paper's white point and view that color, then switch out prints, get fully adapted to the other paper's white point and view its version of that color, the two colors, though they have different XYZs, should "appear" to be relatively the same color, because I was adapted to the different white points.

Keeping Lab values the same between two different white points is in essence using an "XYZ Scaling" Chromatic Adaptation Transform to model the appearance of colors when adapted to the two different white points because that XYZ Scaling CAT is built into the Lab to XYZ and XYZ to Lab transform functions when changing white points.

But we've learned that XYZ Scaling isn't as good in practice as several other, more complex transforms, like Von Kries or Bradford (because they convert first into a cone response space, do the adaptation transform there, then convert back into XYZ.)  So, to model the relative appearance of colors when adapted to different white points, ICC profiles/CMMs use one of those better transforms, and in doing this, we end up not only with different XYZ values for a relative color when adapted to different white points, but also with different Lab values for that color.

[Doug Gray]

Is CIE 1976 Lab just a math transform to and from XYZ that yields a more perceptually uniform space relative to a specific white point, or does that transform also provide a chromatic adaptation function which makes Lab a form of appearance model between different white points?  I've seen it presented both ways and it's confusing.  Here's my current understanding.  Please correct me if I'm wrong.

Yes, and no. L*a*b* (Lab hereafter) always requires a WP reference. So XYZ, given a white point, always converts to and from Lab which is more perceptually even.  ICC profiles specify D50 as the illuminant and a 100% reflective paper. XYZ thus has a defined conversion to/from Lab. As such this ideal paper's white point is also exactly D50. However, profiles are made relative to the paper's actual white point and using Relative Colorimetric Intent will print Lab values relative to that white point with the unprinted media being Lab=(100,0,0).

If you use Absolute Colorimetric Intent a D50 based Lab color will be printed and the paper's white point is irrelevant. Of course the printable gamut no longer extends to Lab(100,0,0) but somewhat lower than the actual paper's white to accommodate the need to override the paper's tint.

Say I have two printer profiles with different paper white points.  They both assume a D50 illuminant is being used to view the prints.  A Lab color inside the first printer/paper space will have it's own specific XYZ value.  If I use those same Lab values in the other printer/paper space, that color will now have different XYZ values because the white point is different.  If I get fully adapted to the first paper's white point and view that color, then switch out prints, get fully adapted to the other paper's white point and view its version of that color, the two colors, though they have different XYZs, should "appear" to be relatively the same color, because I was adapted to the different white points.

Keeping Lab values the same between two different white points is in essence using an "XYZ Scaling" Chromatic Adaptation Transform to model the appearance of colors when adapted to the two different white points because that XYZ Scaling CAT is built into the Lab to XYZ and XYZ to Lab transform functions when changing white points.

Yes, this is how Relative Colorimetric works with different paper whites.

But we've learned that XYZ Scaling isn't as good in practice as several other, more complex transforms, like Von Kries or Bradford (because they convert first into a cone response space, do the adaptation transform there, then convert back into XYZ.)  So, to model the relative appearance of colors when adapted to different white points, ICC profiles/CMMs use one of those better transforms, and in doing this, we end up not only with different XYZ values for a relative color when adapted to different white points, but also with different Lab values for that color.

Since paper white points are typically quite close to D50, the transform introduces almost no error and is rarely perceivable. Also, even the better transforms do not produce perfect metamers. V4 profiles allow for separate Rel Col and Abs Col 3DLUTs which addresses this perfectly but I don't think it is used outside of perhaps scientific work where super precision is needed.

[BradFunkhouser]

even the better transforms do not produce perfect metamers.

It's my understanding that metamers are different spectral power distributions that yield the same XYZ values from the observer functions.  So this is tristimulus response to colors in isolation, colors not coordinated or relative to any other input, having nothing to do with white point adaptation.  Right?

With metamers, you see colors as matching, side by side, viewed simultaneously.  But you can't be adapted to different white points simultaneously, so it seems like we're talking about something slightly different when we're in the appearance model domain.  Is there a different name, analogous to metamers, for colors that "appear" the same when the observer is fully adapted to different white points?  In studying the logic of color management, I've thought for a while that having identical Lab values but different white points served exactly this purpose.  But now I'm thinking that's only partially true, because, for colors that should appear the same when adapted to different white points, the Lab values will actually be different when you start using alternative CATs.

[Doug Gray]

It's my understanding that metamers are different spectral power distributions that yield the same XYZ values from the observer functions.  So this is tristimulus response to colors in isolation, colors not coordinated or relative to any other input, having nothing to do with white point adaptation.  Right?

With metamers, you see colors as matching, side by side, viewed simultaneously.  But you can't be adapted to different white points simultaneously, so it seems like we're talking about something slightly different when we're in the appearance model domain.  Is there a different name, analogous to metamers, for colors that "appear" the same when the observer is fully adapted to different white points?  In studying the logic of color management, I've thought for a while that having identical Lab values but different white points served exactly this purpose.  But now I'm thinking that's only partially true, because, for colors that should appear the same when adapted to different white points, the Lab values will actually be different when you start using alternative CATs.

Sure, metamers are spectrally different but appear the same. When creating printer profiles, the media's white point illuminated with D50, is typically somewhat bluish compared to a perfect diffuser. Thus it has a spectral absorbance that is not uniform and can vary yet result in the same XYZ illuminated by D50. When using the same LUTs (AtoB1, and BtoA1) for Rel Col and Abs Col, the differences are adjusted by purely linear scaling which is less than idea but, because WPs are typically not too far off, are not a significant factor for most media. Still, the spectral absorbance of the media introduces error and is not generally a match for the absorbance characteristics of the inks. The metameric fail (although typically miniscule) will usually come about with different printers printing the same paper and Rel Col Lab value (hence using the same paper white point of Lab=100,0,0) since the profiles are made with D50 Lab readings. As I indicated, these effects are extremely small but are eliminated by using separate tables (AtoB3 and BtoA3) for Abs.

This is, however, negligible compared to major illuminant change CATs.

CATs are necessary approximations because physical objects have variable reflectance spectra. So they are derived using defined sets and minimizing some metric such as mean dE76 or median dE94.

Interesting discussion of some of the issues evaluating CATs.
https://infoscience.epfl.ch/record/52171/files/SusstrunkF05.pdf?version=2

[BradFunkhouser]

Interesting discussion of some of the issues evaluating CATs.
https://infoscience.epfl.ch/record/52171/files/SusstrunkF05.pdf?version=2

Thanks for the link.  Very interesting.  After reading that paper and others, I've tried to work out a demonstration to show what a CAT does based on my understanding.  Please correct me where I'm wrong.
 
To help simplify the scenario, I'm using only emissive colors and their XYZs; no illuminant SPDs or object reflectance SPDs involved. And assuming participants' CMFs match the standard observer.
 
Start with a large field of white. Add two colored lights side by side with the same XYZ values.  They will match in appearance.  Vary that surrounding white and those two colors will continue to have the same XYZs and will continue to match each other, even as your visual system adapts to the different white points.  This shows the persistence of the color matching functions across different viewing conditions.  This persistence is because the CMFs are in essence modeling cone photopigment absorption, the earliest stage of encoding, before the signal levels are adapted or processed for context.
 
Move the colored lights apart and surround each with a different white.  They will no longer appear to match, even though they still have the same XYZ values.  This shows appearance difference due to a change in context (this particular effect is simultaneous contrast).
 
Put a partition between the colored lights that splits the visual field in half so one eye sees its colored light surrounded by one white, and the other eye sees its colored light surrounded by the different white.  After a few minutes, each eye will adapt to its own white.  The two different whites will now appear the same even though they have different XYZs.  This shows sensory level white point adaptation.  The colored lights will appear different, even though they still have the same XYZs, because their appearance has been shifted by the different white point adaptations.
 
Adjust the colored light on one side until a match is perceived with the colored light on the other side.  The adjusted color now has a different XYZ, but the two colors now match in appearance across the different white points.  This is a manual chromatic adaptation for that one color.
 
Given the XYZ of the unadjusted color and of the white that surrounds it, together with the XYZ of the white on the other side of the partition, a CAT should come close to predicting the XYZ of the manually adjusted colored light.

[Doug Gray]

Thanks for the link.  Very interesting.  After reading that paper and others, I've tried to work out a demonstration to show what a CAT does based on my understanding.  Please correct me where I'm wrong.
 
To help simplify the scenario, I'm using only emissive colors and their XYZs; no illuminant SPDs or object reflectance SPDs involved. And assuming participants' CMFs match the standard observer.
 
Start with a large field of white. Add two colored lights side by side with the same XYZ values.  They will match in appearance.  Vary that surrounding white and those two colors will continue to have the same XYZs and will continue to match each other, even as your visual system adapts to the different white points.  This shows the persistence of the color matching functions across different viewing conditions.  This persistence is because the CMFs are in essence modeling cone photopigment absorption, the earliest stage of encoding, before the signal levels are adapted or processed for context.
 
Move the colored lights apart and surround each with a different white.  They will no longer appear to match, even though they still have the same XYZ values.  This shows appearance difference due to a change in context (this particular effect is simultaneous contrast).
 
Put a partition between the colored lights that splits the visual field in half so one eye sees its colored light surrounded by one white, and the other eye sees its colored light surrounded by the different white.  After a few minutes, each eye will adapt to its own white.  The two different whites will now appear the same even though they have different XYZs.  This shows sensory level white point adaptation.  The colored lights will appear different, even though they still have the same XYZs, because their appearance has been shifted by the different white point adaptations.
 
Adjust the colored light on one side until a match is perceived with the colored light on the other side.  The adjusted color now has a different XYZ, but the two colors now match in appearance across the different white points.  This is a manual chromatic adaptation for that one color.
 
Given the XYZ of the unadjusted color and of the white that surrounds it, together with the XYZ of the white on the other side of the partition, a CAT should come close to predicting the XYZ of the manually adjusted colored light.

Interesting, and easily set up experiment but it may show big differences depending on how much adaptation is done in the brain after merging the processed information from each eye. I don't recall reading anything that breaks it down at that level but would be surprised if some work like that hasn't been done. Have you researched the scholarly literature?

[BradFunkhouser]

Interesting, and easily set up experiment but it may show big differences depending on how much adaptation is done in the brain after merging the processed information from each eye. I don't recall reading anything that breaks it down at that level but would be surprised if some work like that hasn't been done. Have you researched the scholarly literature?

Dr. Fairchild mentions testing of chromatic adaptations and appearance models done in this split view way (haploscopic) in his book "Color Appearance Models" (2013) which I read parts of over the holidays.  I haven't dug into the details of the specific studies he mentioned yet. 

I was really intrigued by the potential of being able to see the chromatic adaptation transforms at work in a simultaneous way like that -- it would be such a cool teaching tool!  And in coming up with that scenario I was trying to sort out in my mind that CATs do work with just XYZs of white points and emissive lights (stepping back from thinking about them so much in terms of object spectral reflectances and illuminant SPDs). 

I can't tell from what I've read so far how much factors like incomplete adapation and sensory level versus cognitive level adaptation would come into play for this type of test or how those factors would impact the predictions of the CATs.  Lots to learn.

[Doug Gray]

Dr. Fairchild mentions testing of chromatic adaptations and appearance models done in this split view way (haploscopic) in his book "Color Appearance Models" (2013) which I read parts of over the holidays.  I haven't dug into the details of the specific studies he mentioned yet. 

I was really intrigued by the potential of being able to see the chromatic adaptation transforms at work in a simultaneous way like that -- it would be such a cool teaching tool!  And in coming up with that scenario I was trying to sort out in my mind that CATs do work with just XYZs of white points and emissive lights (stepping back from thinking about them so much in terms of object spectral reflectances and illuminant SPDs). 

I can't tell from what I've read so far how much factors like incomplete adapation and sensory level versus cognitive level adaptation would come into play for this type of test or how those factors would impact the predictions of the CATs.  Lots to learn.

Sound good. I, and probably others, would be interested in any experiments you do.

[BradFunkhouser]

I'm working on that haploscopic matching test as a learning exercise. 

My display white measures D65 at 250 cd/m^2. 

For this image on my display, the left side measures D75 at 200 cd/m^2, the right side measures D55 at 200 cd/m^2. 

The room is dark other than the display.  I position a piece of black foamcore to split my field of view so each eye sees only its half of the display.  After about a minute, those two whites appear nearly identical to me.  Introducing a D65 gray that spans the two sides gives a dramatic sense of the independent adaptations that have occurred.

[Doug Gray]

I'm working on that haploscopic matching test as a learning exercise. 

My display white measures D65 at 250 cd/m^2. 

For this image on my display, the left side measures D75 at 200 cd/m^2, the right side measures D55 at 200 cd/m^2. 

The room is dark other than the display.  I position a piece of black foamcore to split my field of view so each eye sees only its half of the display.  After about a minute, those two whites appear nearly identical to me.  Introducing a D65 gray that spans the two sides gives a dramatic sense of the independent adaptations that have occurred.

interesting. So it appears color adaption is very largely done before the brain starts combining the images from each eye.

[MarkM]

interesting. So it appears color adaption is very largely done before the brain starts combining the images from each eye.

I think it has been established for quite some time that individual cone cells have a photo-chemical gain control that figures prominently into both color and light adaptation. There's also quite a bit of literature that suggests that physical effects in the retina are not enough to explain all phenomena.

There's a paper by Mark Fairchild called "Successive-Ganzfeld Haploscopic Viewing Technique for Color-Appearance Research" (I know that's a mouthful) that discusses different ways of testing with haploscopic setups that might be inspiration for further experiments.

[BradFunkhouser]

Thanks for pointing out that article.  I can certainly attest that the "binocular rivalry" of a traditional simultaneous haploscopic technique is annoying, as the authors point out.  I'd actually already tried what they refer to as the "simple successive" haploscopic technique by alternating closing one eye then the other because I wanted to try to avoid that annoyance.  Using an alternating neutral diffuser (rather than totally covering one eye) to make it a "successive-Ganzfeld" haploscopic technique is very interesting.  According to their findings, it minimizes the loss of adaptation as you switch between eyes and also minimizes the triggering of cognitive mechanisms.

With regards to sensory versus cognitive level adaptation, I found in my simple test that the sensory only adaptation is very fragile.  Raising a hand up barely into view on either side of the partition, so it's lit by just one side of the display and seen by just one eye, causes a partial loss of the adaptation almost instantly.  And that seemed about equally true for either side.  Not very scientific, I know, but it seems like my brain might have some definite ideas about how my hands should look.

[Doug Gray]

Thanks for pointing out that article.  I can certainly attest that the "binocular rivalry" of a traditional simultaneous haploscopic technique is annoying, as the authors point out.  I'd actually already tried what they refer to as the "simple successive" haploscopic technique by alternating closing one eye then the other because I wanted to try to avoid that annoyance.  Using an alternating neutral diffuser (rather than totally covering one eye) to make it a "successive-Ganzfeld" haploscopic technique is very interesting.  According to their findings, it minimizes the loss of adaptation as you switch between eyes and also minimizes the triggering of cognitive mechanisms.

With regards to sensory versus cognitive level adaptation, I found in my simple test that the sensory only adaptation is very fragile.  Raising a hand up barely into view on either side of the partition, so it's lit by just one side of the display and seen by just one eye, causes a partial loss of the adaptation almost instantly.  And that seemed about equally true for either side.  Not very scientific, I know, but it seems like my brain might have some definite ideas about how my hands should look.

Funny you should mention that.

The cognitive aspects of sensing color create some of the most amazing visual effects.

As an example, consider how white paper looks. I typically run my room with about 120 lux of standard incandescent light on the desk of my workstation with a CCT around 3000K. My monitor runs at 100 nits and a D50 white point.

That paper looks mostly white on my desk. Just slightly yellowish relative to my monitor white. But it's a psychological artifact.  Then I measure the actual XYZ values reflected from that paper and then set exactly the same XYZ values on a paper size image on my monitor with a black surround and set to full screen. Does it look the same as the piece of paper on my desktop? No.  Not even close. It looks like a dingy yellowish brown. And the paper still looks "white."

Now, if I carefully lift the paper up and approach the dingy yellow image on my monitor the two continue to look very different until I almost align the paper with the screen image. Then suddenly my perception of the colors change and they look nearly the same. Move the paper away and the perceived difference returns.

It's quite startling and not even close to a subtle effect. The perceptual difference corresponds to about 30 to 40 dE until they get close then the colors jump together within a few dE.

This is obviously a cognitive effect from "knowing" the paper is white but associating the image in the monitor as a dingy yellow.