Color management myths and misinformation video

[Slobodan Blagojevic]

Yes...

So, you are now saying that the sentence I quoted from YOUR quote from "the nice folks at CHROMIX" (post #261) is not correct!? You are now disputing your own quote!?

I can understand that you guys are arguing with each other, but you, Andrew, are now arguing with yourself.

[Alan Goldhammer]

Well, if we agree that color is not a physical property but a perception of our brain, then it cannot be possibly be described by a physical unit. The "volume" in this case is an abstraction.

But it does have a component that can be measured and we do this all the time in science.  Back in the good old days I often used different spectrophotometers to make laboratory measurements in experiments that I was carrying out.  I also often used fluorescent molecules to tag proteins and the instrument made the measurement, not my brain (though I could see the color of the dye visually).

[digitaldog]

So, you are now saying that the sentence I quoted from YOUR quote from "the nice folks at CHROMIX" (post #261) is not correct!? You are now disputing your own quote!?

No, I'm disputing their quote as you misunderstand it.
In your need to disagree to be disagreeable, you've decided to filter all the other points made here and a few even the CHROMIX email states (which is typical of you):

(Now that statement comes with our usual caveat that this volume number is a rough estimate, not a precise one - and it works well for and is intended for making comparisons between profiles, not for defining absolute volume numbers.)
ON the OTHER hand…..There is a philosophical issue at stake here:  Just what constitutes a color?

Further, what CHROMIX is talking about is gamut volume, NOT NUMBER OF COLORS. Please try to read and understand the points made before automatically deciding to disagree, got off topic or attempt to be another color management comedian. You're not very good at any!

[Jim Kasson]

That is the crux of the "does a wider gamut working space have more colors than the smaller one?" Colors have to be humanly-[perceptible] to be considered colors and thus counted.

Andrew, a few posts ago to Sandy I questioned the utility of gamut volume calculations. This whole "number of colors" thing is, IMHO, even less useful. If we're trying to describe the difference between sRGB and Adobe RGB, I think the way to put it is: "Every color you can describe (maybe too technical a word; feel free to substitute) in sRGB you can describe in Adobe RGB. In addition, there are colors describable in Adobe RGB that can't be described in sRGB."

Over the years, I've seen all kinds of problems that resulted from inappropriately (it's fine for comparing emissive displays) looking at color in chromaticity spaces. Although I appreciate the advantage in simplicity gained from going from three dimensions to two, often important information is lost. Trying to reduce color to a scalar is even more dangerous.

Therefore, I think the challenge of the video is to get across the nature of color gamuts in three dimensions to people who are unsophisticated in envisioning 3D objects from looking at 2D displays (paper, computer monitor, TV, etc). When I was earning a living as a color scientist and trying to explain color to people who didn't understand much about it (which was almost all of IBM), i resorted to props. One of the most effective was the Munsell Color Tree. I worked best when people could handle it, but I think that it could be a good tool to use in your video, flipping through the leave and explaining luminance, hue, and chroma. The axes aren't quite the same as the cylindrical version of Lab, but they're close enough to understand the Lab space in a general manner, which is all a neophyte needs to know to appreciate 3D gamut plots.

Then there's how to present 3D gamuts in a video. I find that static presentations are difficult for inexperienced people to appreciate. It's best when they can control the viewing angle themselves, which is not possible in a video, but watching the display as you vary the angle would probably be l lot more effective than a static display.

Jim

[Eric Myrvaagnes]

I have followed the closed thread and this one with considerable interest.
When I first started playing with digital photography, after some 50 years of black-and-white darkroom work, I was thrilled to be able to make "color" prints without having to send slides to Kodak or take color negatives to the corner drug store. I had never heard the phrase "color management" in my life until I had visited LuLa for a while.

Since then, thanks almost entirely to the LuLa resident gurus (Andrew, Jeff, Michael, and Eric Chan) I have learned some of the rudiments, and the results I get with calibrated monitor, reasonable printer profiles, and soft-proofing are quite generally satisfactory.

Until fairly recently I was still a bit unclear about the difference between "assign profile" and "convert to profile."

Thus, I think I qualify as a good target for Andrew's upcoming video, and I look forward to it.  But I must say that this thread has clarified so many things for me that my understanding of Color Management has improved immensely.

So I want to thank you all, especially Andrew, for working so patiently to explain the key points and address the key questions that have developed here. And I guess I even want to thank Slobodan for some of his pesky questions. However, I am certain that Slobodan is no dummy, and I do wish he didn't try to insist that every concept, however complex, be explainable in twenty-five one-syllable words or less. I would love to see him give a 25-word, clear, simple explanation of General Relativity.  

Eric M.

[digitaldog]

I would love to see him give a 25-word, clear, simple explanation of General Relativity.  

Or solipsism.

[digitaldog]

Andrew, a few posts ago to Sandy I questioned the utility of gamut volume calculations. This whole "number of colors" thing is, IMHO, even less useful.

Agreed. That's why several times here I've pasted Graeme's simple statement: therefore it's wrong to talk about number of colors.
And that's an answer to the original myth I hear over and over again I was hoping to address in a video (among other myths), namely this working space has more colors than that working space. And perhaps it doesn't have to be aimed at the beginner.

Therefore, I think the challenge of the video is to get across the nature of color gamuts in three dimensions to people who are unsophisticated in envisioning 3D objects from looking at 2D displays (paper, computer monitor, TV, etc)

Agreed. And that's why a video can work where a paper can't. We can show the gamuts in 3D, move them around etc. The only excuse for showing a 2D gamut is if you're limited to only two dimensions ;-)

[Slobodan Blagojevic]

No, I'm disputing their quote as you misunderstand it.
In your need to disagree to be disagreeable, you've decided to filter all the other points made here and a few even the CHROMIX email states (which is typical of you):
Further, what CHROMIX is talking about is gamut volume, NOT NUMBER OF COLORS. Please try to read and understand the points made before automatically deciding to disagree, got off topic or attempt to be another color management comedian. You're not very good at any!

You quoted them, without saying you disagree. You are only now disagreeing, when caught contradicting yourself.

I am not unaware of their caveat ("that this volume number is a rough estimate, not a precise one") and a rough estimate is good enough for me (and them, apparently). That why neither I nor them use a "number" but a descriptor "more," which again is good enough for me. We do not know how much more is "more," but more it is.

Which also stands to reason: if there is the same number of theoretical colors in each space, it follows that wider spaces would have wider distances between "addressable locations," thus those wider distances would be easier discerned by the eye. The eye would have more trouble discerning different colors in a "squeezed" space like sRGB, then in a wide space like ProPhoto RGB.

[Slobodan Blagojevic]

... I would love to see him give a 25-word, clear, simple explanation of General Relativity.  

Thanks for the opportunity, Eric!

This is how Einstein himself explained: "Imagine one minute spent with a cute secretary... then imagine the same minute spent with your bare ass on a hot stove... which one is longer? That's relativity!"

[digitaldog]

You quoted them, without saying you disagree. You are only now disagreeing, when caught contradicting yourself.

But I don't. I disagree with your understanding of what they wrote. Big difference.

I am not unaware of their caveat ("that this volume number is a rough estimate, not a precise one") and a rough estimate is good enough for me (and them, apparently). That why neither I nor them use a "number" but a descriptor "more," which again is good enough for me. We do not know how much more is "more," but more it is.

Gamut volume isn't number of colors. Nor is it number of encoded numbers of a pixel! Try connecting the dots.

Which also stands to reason: if there is the same number of theoretical colors in each space, it follows that wider spaces would have wider distances between "addressable locations," thus those wider distances would be easier discerned by the eye. The eye would have more trouble discerning different colors in a "squeezed" space like sRGB, then in a wide space like ProPhoto RGB.

It only stands to reason in your mind at this point of understanding the topic. The colorimetric distance between colors in a wider color space is father apart than in a smaller space. Back to the half inflated balloon with 16.7 million dots vs. the balloon twice the same. That doesn't mean (and it isn't true) that there are 16.7 million colors! It's foolish to assume encoding of color values equals colors!
It's got NOTHING to do with "those wider distances would be easier discerned by the eye" as you write. The color is either perceived and counted or it isn't perceived and isn't a color and not counted.

[Slobodan Blagojevic]

But I don't. I disagree with your understanding of what they wrote. Big difference. Gamut volume isn't number of colors. Nor is it number of encoded numbers of a pixel! Try connecting the dots.  It only stands to reason in your mind at this point of understanding the topic. The colorimetric distance between colors in a wider color space is father apart than in a smaller space. Back to the half inflated balloon with 16.7 million dots vs. the balloon twice the same. That doesn't mean (and it isn't true) that there are 16.7 million colors! It's foolish to assume encoding of color values equals colors!

My understanding is exactly the same as their's, i.e., I am OK with a rough estimate, as they are. I did not, nor they did, use the word "number," just a more general descriptor "more," which again is fine by me.

Thus their sentence "the larger gamut will necessarily represent more distinct, humanly-perceptable colors than a smaller gamut." I accept as correct.

Now, by you own admission, you are not good at math, nor you are a scientist, and it appears now that you are not good at formal logic as well.

[digitaldog]

My understanding is exactly the same as their's, i.e., I am OK with a rough estimate, as they are. I did not, nor they did, use the word "number," just a more general descriptor "more," which again is fine by me.

It has nothing to do with rough or fine estimates, you still don't get it. Your understanding isn't exactly the same as theirs. That's the problem.

Thus their sentence "the larger gamut will necessarily represent more distinct, humanly-perceptable colors than a smaller gamut." I accept as correct.

You accept it not understanding much of what has transpired over the last pages or just don't want to accept them. You accept it by ignoring what was written in the same email:

it works well for and is intended for making comparisons between profiles, not for defining absolute volume numbers
Depending on how the numbers are encoded, you can have several million combinations of different numbers representing different colors in theory, but some will point out that these different number combinations do not constitute individual “colors” since they are not distinct enough to be *different* to the human vision system.

You continue to ignore what the gamut volume metric is designed to describe. Believe whatever you wish.

[Alan Goldhammer]

I would love to see him give a 25-word, clear, simple explanation of General Relativity.  

An apple falling from a tree hits Newton's head but not Einstein's.

Did it in 12 words!!!

[Slobodan Blagojevic]

As a "man of science" Andrew, even if it is faux science in your case, you should be familiar with the ceteris paribus principle in reasoning (i.e. "other things being equal"). You keep altering "other things" when they should remain equal. You keep introducing non-discernable colors into a discussion about the discernible ones. I am glad you recently learned about a logical fallacy  known as "straw-man argument" as your reasoning is full of those. Start applying what you recently learned.

[MarkM]

- Could it be theoretically possible that we could discern more colors in a wider space, given that "the larger gamut will necessarily represent more distinct, humanly-perceptable colors than a smaller gamut."

I think you got the crux of the problem right here Slodoban.

If you can define individual humanly-perceptable colors in terms of a closed volume in LAB space, then the answer has to be yes. The volume doesn't even have to be uniformly-sized throughout the space, it just needs a clear boundary and a defined volume. If you do that and you have a positive, defined volume in LAB space for individual colors, then clearly a larger volume will, in theory, hold more of these. You can even go through the trouble of not counting these units when they fall outside the spectrum.

The problem is that it is quite difficult to define these volumes of humanly-perceptable colors in LAB space. Although it's in the right spirit, ∆E doesn't work.

The first problem you run into when trying to define these volumes is one of definition. We've decided that a unique color needs to be discernibly different. So even if the numbers are different, they don't count as uniques colors if they look the identical, right?. So this would count as one color even though the image is made of two color values:



But if we are going to be strict and define colors by human perception, then how many greens are in this image?



The numbers say one, but my eye says more than one. So how many do we count it? Regardless of how you defined the volumes above, all these green will fit on one point in LAB space, yet by our own definition there is more than one color here. That's a paradox that we can't solve using our current colorimetry (although research on color appearance models is working on it).

You now might be tempted to change your definition of color to exclude the messy realities like simultaneous contrast and decided matches under laboratory conditions with set surrounds and white points, but then your volume calculations only work in the laboratory and you're back in a lot of ways to where you started.

[digitaldog]

You keep introducing non-discernable colors into a discussion about the discernible ones.

And you keep taking a simple statement you don't really understand and trying to use it to prove something that doesn't make sense.
Take the statement of gamut volume from CHROMIX with respect to ProPhoto RGB while considering the portion of that color space (the blue primary) analyzed which falls outside human vision. Does ProPhoto RGB have a larger gamut volume in that respect?

[Slobodan Blagojevic]

I think you got the crux of the problem right here Slobodan...

Thanks.

Mark, I understand the complexity of the issue, and particularly when you introduce an optical illusion, as you did. I understand the difficulty of measuring something very precisely. However, as I said, sometimes a rough estimate in the right direction is enough for me and for all practical purposes. There are many things in science that are intuitively arrived to, in absence of measuring. For instance, Einstein's theory of relativity only recently got a confirmation through measurement.

[digitaldog]

The problem is that it is quite difficult to define these volumes of humanly-perceptable colors in LAB space. Although it's in the right spirit, ∆E doesn't work.

Exactly. And let's not forget (or let Slobodan forget) how Lab and thus dE or the original mathematical models of which it's based come about in the first place. How the CIE came up with CIE XYZ (1931) to build the model in the first place. The tests involved showing groups of volunteers a sample color under very controlled conditions whereby each subject adjusted the intensity of red, green, and blue lights until the mix of the three matched the sample color.
One of the best and well explained posts on this subject came from Fred Bunting who wrote the original treatise on color management for the ColorTron years and years ago. I've kept this post from the CS list dated back in 1998 because it is so well written:

Subject:   Re: Understanding Colour Gamuts
Date Received:   Wednesday, November 25, 1998 2:06:53 PM
Date Sent:   Wednesday, November 25, 1998 2:06:53 PM
From:   Fred Bunting <fred_bunting@ls.com>
Cameron writes ... a lot of questions.  Most of this is addressed in the
ColorShop Color Primer which I wrote but is currently available only as
part of the documentation for various X-Rite products <shameless plug>.
But I'll 'summarize'.

Actually, I am amazed at how well you seem to have 'guessed' the right
answers to most of your questions.

The XYZ space is derived from color matching studies using human test
volunteers.  From this data, the CIE defined three imaginary primaries X,
Y, and Z that model the color matching characteristics of the average test
subject, (what the CIE calls the Standard Observer).  By 'imaginary' I mean
that these primaries do not actually exist in nature, but are
mathematically derived from the experimental data.  And by 'model' I mean
that we can predict when most people would consider two stimuli a 'match',
based on whether the two stimuli have the same X, Y, and Z values, i.e.
whether the two stimuli have the same location in XYZ space.

And yes, you are correct that this matching must be done under some
reference lighting condition.  A light source defined spectrally (i.e. in
terms of the energy at each wavelength of the spectrum) is called an
'illuminant'.  The CIE has outlined a system of Standard Illuminants,
labeled A, B, C, D, E, and F.  The D and F illuminants are actually series
of illuminants, representing various 'daylight' and 'fluorescent'
illuminants respectively.  The D illuminants include D50, D55, D65, and
D75.  The number represents the 'correlated color temperature' (which, yes,
does vary for daylight depending on time of day and latitude).  By far the
most commonly used illuminants are D50 and D65, corresponding approximately
to 5000K and 6500K respectively.

The large asymmetrical diagram you speak of is indeed a 2D plot of the CIE
XYZ color space.  More correctly, it is a projection of the space onto a
plane.  The axes of this plane are labeled simply (lower-case) x and y and
and the x and y values for a given color are known as the 'chromaticity
coordinates' for that color.  The 'achromatic' (luminance) component of a
color can be represented simply by the (cap) Y value from XYZ ... another
clever feature of the way the CIE defined the X, Y, Z, primaries.  Thus xyY
is often thought of as another 'space' for representing colors.

All of the above (XYZ, the xy-diagram, Standard Observer, and Standard
Illuminants) were all defined by the CIE in 1931.

CIELAB was 'recommended' by the CIE in 1976, to address a specific problem:
namely, while identical XYZ values could tell you when two stimuli would be
experienced as the same 'color' by most observers, it did not tell you how
'close' two colors were if they were not exactly the same XYZ.  I.e. if you
wanted to know whether sample A was more closely approximated by sample B
or C, the distances between the three points, A, B, and C in XYZ space did
not give the right answer as judged by test subjects.  A number of
industries were asking for a way to compute 'color distance'.  To this end,
the CIE introduced L*a*b* (otherwise written, CIELAB), which was just a
mathematical warping of the XYZ space to make the distances correlate
better with the results of test subjects.

CIELUV, by the way, was also introduced in 1976, for essentially the same
purpose, and was a rival method of warping the space ... and the two spaces
developed their different adherents in different industries.

The purpose of both CIELAB and CIELUV was to define a color difference
metric, known as 'delta-E' that is simply the distance between two points
in the color space (in this case CIELAB or CIELUV, although several other
ways of computing delta-E have also been proposed both within and outside
of the CIE, since 1976).

That's the general scoop, now to specific questions:

I vaguely remember someone on this list telling us that some or all CIELAB steps were not equal and I now want to know why?

CIELAB was an *attempt* to create a space where equal steps correlated to
color 'closeness' as judged by test subjects.  No one ... especially not
the CIE ... claims that CIELAB is perfect in this regard.  For example,
most color scientists will tell you that CIELAB tends to overexaggerate
distances in the yellows, and underreport distances in the blues.  It also
tends to 'bend hue lines', i.e. if you try to increase saturation only, by
moving directly away from white in CIELAB space, most test subjects
perceive a change in hue as well ... again most pronounced in the blues ...
which is why we often see the classic purple sky problem.  People have to
remember that CIELAB was invented in 1976 long before its purpose as a
computational space for 1990s digital Color Management was even a dream.
As such it has done remarkably well, but it is not perfect, and this is a
big area of research both by color scientists, and companies working in
color management.

I am told that the eye sees different wavelengths and the mind see them as
different colours.

May I wax philosophic?  One legacy we have from Mssr. Descartes is to
separate things where no separation really exists in nature.  To me there
is no dividing line between the 'eye' and the 'brain' ... color
'processing' occurs right in the first three layers of neurons in the
retina, and right down optic nerve to various way-stations in the brain on
the way back to the visual cortex ... and processing doesn't stop there.
Whether there is a dividing line between the 'brain' and the 'mind' (as you
put it) is the classic mind-body problem through which Descartes (IMO)
completely screwed us up. ...  But that's another thread.  :-)

Does the reference light source contain equal  `amounts` of each wavelength,

That is illuminant E, the 'equal energy' illuminant ... entirely
hypothetical as no such illuminant actually exists in nature.  This is just
one of several reference illuminants.

Do the three cones in the eye respond to differning wavelengths or wavebands in the same or different ways, with
equal or varying efficiency?

Yes, each type of cone responds to different regions of the spectrum, and
each has a peak wavelength that it is most responsive to with this response
trailing off in both directions up and down the spectrum.

In short, what accounts for the asymetry in the diagram, is it based in reality ...

It (the assymetry of the xy-diagram) is based on mathematical manipulation
of experimental data.  I.e. it is indeed based on psychophyics (measuring
psychological response to a physical stimulus), but you should not read too
much into the *degree* of assymetry, as this may be exaggerated (or
diminished) by the mathematics.

... and is it [this assymetry] necessary for colour management?

Well, the *assymetry* of the xy-diagram is not necessary for color
management ... in fact it is a hindrance that the distances are distorted
... which is why most color management computations use CIELAB which
addresses this distortion.  However, there are other features of the
diagram, such as the fact that it illustrates additive color quite nicely,
and offers a map of the entire visual system, which are useful in
explaining certain things like device gamuts.

Fred Bunting
X-Rite CTC

Well done Fred, well done!

[MarkM]

I agree Slobodan. But before we make the estimate, especially if it's for practical purposes, I would like to know if the question even really makes sense. Since I'm not even sure it's possible to define a finite set of unique colors in terms of LAB volumes in any real-world sense, I'm not ready to entertain the estimate, even in theory until I understand the problem better.

We can certainly choose to ignore the messiness and ask conditionally — if we can define volumes in LAB space that represent unique colors, can we define how many of these sRGB holds and compare it to AdobeRGB? The answer is obviously yes. But the condition such an important part of the original question that ignoring it makes the theoretical answer of little value.

If we're being truly practical about this, the whole question is not really important. The bit depth of an actual image and the range of possible colors is of much more concern. Nobody counts individual number of colors in a file as a function of gamut volume, but images DO posterize due to lack of bit depth and DO clip due to gamut range limitations.

[MarkM]

Well done Fred, well done!

Indeed, that's a really good summary.