attention color whizes: non-typical sRGB/RGB/ProPhoto question

[MarkM]

XYZ space does not have a varying notion of white point

Which is why the spec tells you to use PCS. Which is also why your plot doesn't prove anything and isn't relevant.

Look at it this way. Suppose you have a color patch that you would like to measure and you want to determine the coordinates of the color in XYZ space. One way to do this is measure the color with a photospectrometer. The spectrometer will generate a spectral power distribution of the light reflecting off the patch. You can combine the SPD with the 1931 observer data and get the XYZ coordinates. It's not very hard. But here's a very important question: what color light are you going to shine on that patch to build your SPD? This is very relevant if you want to share the XYZ number with someone else and have them be able to reproduce your color, which is why presumably we are doing all this. Depending on the illuminant you will get very different XYZ numbers, Which one will you use to represent the color of your patch?

[joofa]

Which is why the spec tells you to use PCS. Which is also why your plot doesn't prove anything and isn't relevant.

You sure about it? I don't think that you have even realized what the plot is telling you.

Look at it this way. Suppose you have a color patch that you would like to measure and you want to determine the coordinates of the color in XYZ space. One way to do this is measure the color with a photospectrometer. The spectrometer will generate a spectral power distribution of the light reflecting off the patch. You can combine the SPD with the 1931 observer data and get the XYZ coordinates. It's not very hard. But here's a very important question: what color light are you going to shine on that patch to build your SPD? This is very relevant if you want to share the XYZ number with someone else and have them be able to reproduce your color, which is why presumably we are doing all this. Depending on the illuminant you will get very different XYZ numbers, Which one will you use to represent the color of your patch?

The reflected color off a patch is a product of spectral illuminant and patch's reflectivity. The product is not a good way to determine the "color" of an object/patch, because as you said, it depends upon illuminant. One way to have (kind of) an invariant specification is if you can figure out the patch's reflectivity coefficient.

Joofa

[MarkM]

One way to have (kind of) an invariant specification is if you can figure out the patch's reflectivity coefficient

OK. Show me how to get from a reflectivity coefficient to XYZ without an illuminant.

[ejmartin]

Because to do a representation of a color with tristimulus XYZ=[0.188185   0.075274   0.991108], which incidently happens to be the blue primary of Adobe RGB (D65), in Prophoto RGB (D50), you don't need chromatic adaption. Chromatic adaption is done when there is a question of "what would have been the XYZ values if I had shone D50 on my target surface instead of D65, while having D65 tristimulus values?" That is not at all an issue here.

Sincerely,

Joofa

So it's a purely theoretical question, rather than something one need worry about in an image processing context...

[joofa]

So it's a purely theoretical question, rather than something one need worry about in an image processing context...

Nope. I think Iliah Borg summed it very nicely when he said something to the effect of chromatic adaptation being abused, sometimes not applied when needed, and sometimes applied twice. I am quite surprised to see a confusion on this topic among some of those people who sign their names as "color experts" here (not you of course,  ).

Joofa

[ejmartin]

As long as there are all these "color whizes" here , I have a question: Why when going between illuminants of different color temperature (D65 to D50 for example) one does chromatic adaptation; yet when one does white balance in a raw converter to adjust to the color temp of the illuminant, one multiplies the CFA raw data by some scalar multipliers. They are not the same thing, if I understand correctly.

[digitaldog]

Nope. I think Iliah Borg summed it very nicely when he said something to the effect of chromatic adaptation being abused, sometimes not applied when needed, and sometimes applied twice.

I don’t think anyone would disagree that when chromatic adaptation is needed and isn’t applied (or applied twice) its being abused. That’s not the same as saying chromatic adaptation isn’t necessary nor in this context, should not be applied.

I am quite surprised to see a confusion on this topic...

There’s nothing confusing about applying the use (or lack of in this case) of chromatic adaptation when called for to prove a result you desire.

[Farmer]

Wait.  People are surprised that when you shine a blue light on something it looks different to when you shine a red light on it?  That's essentially what is being done here.  Change from D65 to D50 - might as well be changing from blue to red.

Adobe RGB (1998) is what it is.  If you change it then it's no longer Adobe RGB (1998), which means it's incorrect to compare "it" (the variation) with something to counter a comparison made with the "original" version.

You might as well move both spaces to D75  or D41 or whatever, and then compare - your findings will be as pointless.

For anyone worried about "real world" application, I suggest that you do this.  Find the most saturated blue thing you can and photograph it.  Process the photo using ProPhoto and then using Adobe RGB (1998).  Print the results of each (or display or project or whatever).  See if you get clipping more or less in one or the other.

[Iliah]

That multiplication is often referred to as a form of chromatic adaptation. Things are much more complicated however - still this form of chromatic adaptation results in acceptable balance around midtones.

Under certain types of lighting one does not want to apply full adaptation to the scene, and a mix of relative and absolute is used. For some scenes adaptation is applied differently to different parts of scene, like to shadows and highlights it is applied not in the same manner as to midtones - the "mix" is different.

[MarkM]

2.) Whereas with AbsCol, Adobe’s D65 white should be kept right where it is on the D65 point. At least some CMMs still do so, and it seems to me intuitively correct according to the term "absolute colorimetric".  No chromatic adaptation desired or needed here.  Without chromatic adaptation, D65 white is a non-neutral color in the D50 target space (actually it is even clipped).

Peter, this has been bugging me all day, but I've only now had a chance to look at the spec and figure out why. First, here's why it bugs me:

This is the ICC spec, where they define relative and absolute intents:

A.3.1.2 Media-Relative Colorimetric Intent
This intent rescales the in-gamut, chromatically adapted tristimulus values such that the white point of the actual medium is mapped to the white point of the reference medium (for either input or output).

A.3.1.3 ICC-Absolute Colorimetric Intent
For this intent, the chromatically adapted tristimulus values of the in-gamut colors are unchanged.

In the context of this thread this is confusing and I just realized that it's because we are talking about working spaces where the white point happens to also be the white point of the illuminant. This isn't always the case, in fact for something like a printer profile it would almost never be the case.

For example my epson 3880 profile for one of the papers has a white point of [.909, .943, .793]. The profile also has a Chromatic adaptation matrix to use when moving into PCS. (It is just diagonal 1.0 so it does nothing because the profile illuminant happens to be D50.) So when we are in PCS space if we want relative colorimetric, we scale the media white point to our output white. If we want absolute we leave it and everything else where it is which means on a monitor or proof the whites will look yellow which is what we want for proofing in situations where we want to preserve paper color.

Now this is where it starts to get a little confusing. The illuminant in the printer profile doesn't have to be D50 (although in practice the mostly are). If my profile had a D65 illuminant instead, it should also have a chromatic adaptation matrix that tells us how to get to D50. (The version 2 ICC spec doesn't require this which has been a problem, in version 4 it's required.) The spec tells us that we need to chromatically adapt our tristimulus values and the media white point to PCS (D50). This doesn't mean the white point of my paper will now equal the white point of D50, it just means that the white point has been adjusted for a different chromatic adaptation state—it will still be warm. The new XYZ numbers just represent how the paper would look under D50 light with eyes adjusted for D50 instead of how it looked under D65 light with eyes adjusted for D65—which is to say the same. If we don't do this step we are working with XYZ numbers representing how the paper looks under D65 light with eyes adjusted for D50. We still have our paper white and it is accurately represented in PCS. Now we can either use it to scale a relative intent or not if we want absolute.

Things get screwy when we talk about working spaces because the white point is the same as the illuminant white point. When we perform the chromatic adaptation per the spec the white points of the two illuminants align and relative and absolute colorimetric rendering are the same. They become meaningless. I'm not sure how other CMMs handle this, but Adobe's CMM seems to confirm this.

This doesn't seem as screwy when you think about what should happen with two papers, A and B, that are identical but you've profiled one with a D65 illuminant and one with D50. The paper whites look the same when they are viewed together and they should look more or less the same when viewed separately under D50 and D65 assuming your eyes have adjust to the illuminant. Now if you make two proofs using each profile and absolute colorimetric rendering, one should expect the proofs to look the same. It seems a little counter-intuitive, but the reason is that the profile is not trying to proof the white point of the illuminant in the profile, it is trying to proof the white color of the paper which is by definition is the same in this case. This works if you follow the ICC specification because the white point in the D65 profile goes through the chromatic adaptation on its way to PCS. In PCS space we can see that regardless of the illuminant used to build the profile, the papers are the same color—they'll have the same XYZ numbers for the same viewing conditions.

That's kind of the long explanation, but I think it ties up some loose ends.

[John McDermott]

Obviously I am not expert on any of these issues, but it does seem to me that this thread has become an argument about how many angels are on the head of  pin.

[bjanes]

One way to do this is measure the color with a photospectrometer. The spectrometer will generate a spectral power distribution of the light reflecting off the patch. You can combine the SPD with the 1931 observer data and get the XYZ coordinates. It's not very hard. But here's a very important question: what color light are you going to shine on that patch to build your SPD?

It would seem to me that one should compare the SPD of the illuminant with the SPD of the reflected light. If the illuminant is a continuous function (such as a black body radiator), one could then construct the appearance for any illuminant, including the XYZ wavelengths. This would involve integration across the visible spectrum.

Regards,

Bill

[jc1]

Is gamut of Adobe RGB Fully Enclosed by ProPhoto?  Interesting tropic!
I am here to learn and followings are my observation.

Gamut Edge Data
From a different perspective, I examine it with a set of self generated test points. I use the gamut tool in PatchTool to build a RGB edge list, with a pitch of 5 between RGB value. This pitch is sufficiently small to ensure that generated data points are close enough for showing clearly the gamut boundary for both aRGB and ProPhoto RGB.

The generated RGB data (DS0) consists of total 15,608 points which span across the whole RGB gamut boundary. By assigning the data to aRGB and ProPhoto respectively, 3 sets of data  were derived. The data were output to xyz and read into CTP3.
Data set 1 (DS1): ProPhoto with illuminant D50 (=  ProPhoto)
Data set 2 (DS2): aRGB with illuminant D50 (= Adobe RGB 1998 but shift illuminant from D65 to D50)
Data set 3 (DS3): aRGB with illuminant D65 (= Adobe RGB 1998)

The 3 sets of data were read with CTP3. Followings illustrate their data points in Lab co-ordinate.





Is my data sets tally with the Gamut Plots?
For ProPhoto, the edged data matched nicely with the standard Prophoto gamut, as both are normalised to D50 illuminant.



When displaying with CTP3 (ColorThink Pro 3), the gamut (vector) plot for Adobe RGB 1998 was normalised to D50 illuminant. I got email exchange with  Danny Pascale @ BabelColor,  and he assured me there is nothing wrong with the PatchTool output. I was convinced.




The Verdict
Let the data speak for itself.




sincerely,
jc1

[jc1]

Here's my attempt to construct a true aRGB (1998) profile, based upon 5832 data points and my earlier thought. It was built with a printer profiler, for the seek of displaying the gamut of a True Adobe RGB (1998) @D65 illuminant.





regards

[joofa]

Thanks Jc1 for the interesting plots. It does seem like that Adobe RGB (D65) is going outside Prophoto RGB (D50) gamut in a section from saturated blues to white. Right?

Sincerely,

Joofa

[jc1]

Thanks Jc1 for the interesting plots. It does seem like that Adobe RGB (D65) is going outside Prophoto RGB (D50) gamut in a section from saturated blues to white. Right?

Sincerely,

Joofa

Unless someone else prove it otherwise. 

with regards,
jc1

[sandymc]

I was trying to go back even a bit farther in the evolution of the transform so to speak. When one says ACR mapping is camera to XYZ, what is “camera”? Is some assumption made, are known spectral sensitivities of the camera being used? If the later, are the “colors” imaginary to us excluded?

Andrew,

The known spectral sensitivities are used; Adobe create matrixes for every camera that's supported by ACR/Lightroom. Colors are clipped to the ProPhoto gamut, but given the size of ProPhoto, in practical terms any color that a real visible light camera can record is captured.

Sandy

[sandymc]

Why when going between illuminants of different color temperature (D65 to D50 for example) one does chromatic adaptation; yet when one does white balance in a raw converter to adjust to the color temp of the illuminant, one multiplies the CFA raw data by some scalar multipliers. They are not the same thing, if I understand correctly.

I don't know about other raw converters, but with ACR the entire 3x3 conversion matrix changes.

Sandy

[MarkM]

jc1, you did a  nice job on the the graphs, they're cool looking. The conclusion you draw suffers from the same problem Joofa has been having the whole time, though. If I understand what you've done, you are drawing conclusions about points on the same XYZ plot representing two different viewing conditions.

Nobody has had any luck explaining this to Joofa, but I'll try to start from the very beginning and see if it makes sense to you.

The beginning:
1. XYZ space comes from color matching experiments. XYZ by itself doesn't specify a white point or make any assumptions about the viewing conditions. All it really says is that if you have two points in XYZ that are the same, viewed under the same conditions, the colors will match. Lets call that rule number one. That seems obvious, but it's important because you can have colors that have different spectral components that end up being the same XYZ point (metamers). The converse is true too: if two points are different, they don't match under the same viewing conditions.

I don't think anyone disagrees with the above paragraph.

2. So what about when the viewing conditions are different? Specifically when your eyes are adapted to different light sources. Will the color represented by this XYZ point [.75, .5, .30] look the same when your eyes are adapted to a D65 source as it will when your eyes are adapted ti D50? The answer is no, that color under these two different conditions does not match because you are breaking rule number one. You see this all the time in real life life when you view a tungsten-lit window standing outside at dusk. To you, with eyes adapted to the blue evening light, the light looks very yellow. To people inside, it looks white. You'd get the same XYZ point for it, but it's a different color under different chromatic adaptations.

I don't think anyone disagrees with the above either

3. So what about two different XYZ values each viewed under different conditions? Take our original color [.75, .5, .30] viewed with eyes adapted for D65 and this color [0.78, 0.51, 0.23] viewed under D50. Do they match? You can't tell just by looking at the numbers. If you plot them on a 3D chart they are in different spots in the space and you will be tempted to say they don't match. But they do. Color 1 under D65 matches color 2 under D50.

I don't think anyone disagrees with the above either

4. So now suppose you want to compare two colors, under two viewing conditions on the same graph in a way that makes it clear whether the colors match and not. This is what chromatic adaptation is for. You can take you D65 XYZ values, run them through a chromatic adaptation matrix and get a new value that allows you to compare it to the the D50 value to see if it matches or not. Now when you plot them on the same graph you can see real color relationships, you can make judgments about them. If you don't do this you can't—you can't even tell by looking if two XYZ points match or not if you don't account for different viewing conditions.

So when you plot your AdobeRGB gamut assuming chromatic adaptation to D65 on the same chart as ProPhoto adapted to D50 you can't make any judgments about the relationship between the two sets of colors. You've forgot rule number one. Consider the colors in ARGB that look like they are out of gamut—you can't tell that by looking if they match colors in the proPhoto space or not and hence you can't tell if they are out of gamut or not in the same way that you couldn't tell if the colors in no. 3. above matched.

The point of comparing gamuts is to answer the question: if you give me a color from AdobeRGB can I give you a color from ProPhoto that matches it? Can I do that for all colors in AdobeRGB? In order to answer this and tell if the colors match we need to remember rule number one.

If you want to compare these spaces in a meaningful way you need to account for the difference in chromatic adaptation. If you do, all the problems go away and your results will match the established results. To do otherwise is like me standing outside your house insisting that your lights are yellow and you insisting that the are white. If you want to really talk about the color of the light, either I need to go inside, or you need to come outside.

[jc1]

Hi joofa,

But ...just to clear my stand.
MarkM has made his point loud and clear which I am aware and agree with. 

CIE XYZ space only guarantees that colors with the same coordinates will match when viewed under the same conditions. When you plot various XYZ values in that space taken under different assumptions, like illuminant, you can't expect to be able to compare them. Plotting them on the same graph becomes meaningless. 

regards,
jc1