Color management myths and misinformation video

[Slobodan Blagojevic]

...I'm still working out the details in my brain!

Just make sure you process properly the following from Steve:

"AdobeRGB can hold MORE *unique* color values... the number of distinct colors (as per human vision)... than sRGB"

Works for me, though.

[digitaldog]

Just make sure you process properly the following from Steve:
"AdobeRGB can hold MORE *unique* color values... the number of distinct colors (as per human vision)... than sRGB"
Works for me, though.

I'm sure it works for you, I'm not buying into it yet, I'm still talking with Steve about this. Case in point.


SU: (asking about the two sRGB values that CT reports as having less than 1dE): what are the original color values referring to? are they sRGB or…. and how are they mapped to aRGB?

Andrew Rodney: Yes, sRGB. In 8-bit per color or 16-bit per color, both are less than half a dE as reported by CT. Then when converted to Adobe RGB (RelCol as it was done in Photoshop), they map to the same color value. But let's just stick to the sRGB values. If as CT reports, they are less than 1dE apart, are they two device colors? The dE report would suggest otherwise.

SU: well, colorimetry is human perception in numeric form. That’s kinda it’s basic definition.

Andrew Rodney:But what about color values that appear the same to us? As in the sRGB example above?

SU:I’m confused by the term color values here. If you mean that two device values appear the same then I’d say they’re the same color (unique color in ColorThink parlance). That said, they may be two different color (Lab) values but be too close to tell apart.

Andrew Rodney: I'm confused too because you said: - color values refer to human perception and specifically to colorimetry.
Perception is key isn't it? How can two values that appear the same be two different values based on human perception? They appear the same. See the issues and confusion? IF we have two values, let's call them Color Values, don't they both have to be perceivably different?
-----

See the confusion Slobodan and how we can't take this so easily when CT is saying Color Values refer to human perception but presumably counts color values that we can't tell apart?

[digitaldog]

Here's what I just wrote to Steve.

On Sep 4, 2014, at 5:13 PM, Steve Upton <upton@chromix.com> wrote:

I’d recommend moving from "color value like R89/B87/B255” as you mention above to “device value” - but again that’s my taste.

I'm OK with that once I get all the various flavors figured out. For example, if we are talking about an RGB working space and values, R89/B87/B255 as the example, we can call that a device value?  Considering they are based on devices, that's probably kosher.

It would be useful to have a group of names with color behind them to define what we mean.

Color alone I suspect should be what we perceive.

Device Value as you point out has something to do with a device (and based on my question above, perhaps a working space).

Color Values as you point out refer to perception and specifically to colorimetry.

The confusion is what do we call a value, device or otherwise that isn't perceivable? Color Values doesn't seem kosher if again, the definition is based on perception.

Andrew

I do like what Steve suggests, using Device Value for, well devices. I'm still not clear about Color Values however, certainly not values we can't see as a color.

[Jim Kasson]

I do like what Steve suggests, using Device Value for, well devices. I'm still not clear about Color Values however, certainly not values we can't see as a color.

I don't think we should use Device Value for colorimetric working spaces, even if they are nominally based on (idealized) physical devices. I think that term, if we don't use Colorant Value, should be reserved for the data sent to physical devices or their drivers.

Jim

[Jim Kasson]

I do like what Steve suggests, using Device Value for, well devices. I'm still not clear about Color Values however, certainly not values we can't see as a color.

If we hold the values in a color-managed image to the standard that they all have to represent visible colors, we're going to tie ourselves in knots. Let's see, some of the values in this PPRGB image are color values, and some aren't.  Even stranger: some of the values in this CIEL*a*b* image are color values, and some aren't. Is that helpful? I see your point, Andrew, but maybe a more inclusive term is the answer.

Jim

[digitaldog]

I don't think we should use Device Value for colorimetric working spaces, even if they are nominally based on (idealized) physical devices. I think that term, if we don't use Colorant Value, should be reserved for the data sent to physical devices or their drivers.

I'm OK since they are based on a device (a theoretical one) and I'd hate to have to use another term just for those cases.
Here's more from Steve:

Andrew Rodney:The confusion is what do we call a value, device or otherwise that isn't perceivable? Color Values doesn't seem kosher if again, the definition is based on perception.

SU: well, that’s another good point but I think it’s really just a side effect of assigning numbers to colors.

As soon as you do that you realize that numbers have a habit of going on forever and color perception certainly doesn’t. So we end up with a 3D system where a bunch of the numbers clustered near the center correspond to human perception and if you go too far out then they no longer do.

I think it really only comes up in our world as a result of ProPhoto having an imaginary blue value. It makes sense for the reason that it was used but it does mean that it should probably only be used for advanced users. Certain things, like making a synthetic gradient from 0,0,255 to 255,255,255 means that some of the gradient is “outside human gamut” and should not be expected to look right or convert to other spaces in a pleasing way. Those who expect it to are probably not ready for ProPhoto.

Oh, and also, ColorThink does *not* take this into consideration when calculating gamut volume. It could if we calculated the portion of the ProPhoto gamut that was outside of human perception and subtracted it from the total volume. BUT that would require that we had good, solid data on the edges of human perception and that’s not something I’ve been able to track down either….

Steve

What Steve is saying is what Graeme said I believe: This is a side effect of assigning numbers to colors.
So I still think the question "does Adobe RGB have more colors than sRGB" can't be answered and it's the wrong question. If we talk about device values, they are the same, Steve has said that. Nearly everyone here agrees on that as well. So Adobe RGB and sRGB have the same number of colors (based on encoding) or using Steve's usage, they have the same device values.

[Jim Kasson]

I'm OK since they are based on a device (a theoretical one) and I'd hate to have to use another term just for those cases.
Here's more from Steve:
What Steve is saying is what Graeme said I believe: This is a side effect of assigning numbers to colors.
So I still think the question "does Adobe RGB have more colors than sRGB" can't be answered and it's the wrong question. If we talk about device values, they are the same, Steve has said that. Nearly everyone here agrees on that as well. So Adobe RGB and sRGB have the same number of colors (based on encoding) or using Steve's usage, they have the same device values.

Amen to the "can't be answered and it's the wrong question" part.

However, I've got a problem with this: "It could if we calculated the portion of the ProPhoto gamut that was outside of human perception and subtracted it from the total volume. BUT that would require that we had good, solid data on the edges of human perception and that’s not something I’ve been able to track down either…."

CIE colorimetry says the spectral horseshoe marks that limit. That boundary is buried in Lab, but explicit in Luv. If you're having a hard time finding it in your chosen color space, conversion to Luv should be dispositive.

Jim

[MarkM]

This is a side effect of assigning numbers to colors.

I've mentioned it in passing, and it seems nitpick-y, but in context I think it's worth saying again. We are not assigning numbers to colors. We are assigning numbers to stimuli. That's why the numbers are often referred to as tristimulus values.

It's a very important distinction. We have all this light bouncing around at various wavelengths. We can measure this stimulus with tools that allow much finer precision than we can see. We can take these measurements, manipulate them with equations, then send the numbers to a device that will output some other stimulus. Colorimetry will tell us if these two stimuli — the input and the output — should match under certain conditions even when the spectral components are not the same. If we call these what they are: stimulus values or tristimulus values in the specific case, we will be in agreement with the existing literature and avoid the complications of calling them colors. Colors (as defined by subjective experience) can't be measured, color stimuli can. There's nothing complicated or philosophically problematic with any of this. The stimuli are properties of the outside world, which we can measure and reproduce with as much precision as current technology allows. This is reflected in the continuous nature of the math. The problems only happen when you conflate color stimuli with the subjective internal experience of perceiving color. If you don't do that, the problems go away and the answer to how many colors can AdobeRGB contain is: AdobeRGB doesn't contain colors — it is a measure of color stimulus which is continuous, not discrete.

You can go on from here to discuss the relationship between the stimuli we measure and the experience of perceiving it, but that's a whole different ballgame.

[fdisilvestro]

So when somebody talks about Lch they are just talking about the transforms from LAB we find in sources like Wyszecki (i.e. h = arctan(a*/b*) ) etc.? These calculations just describe the same points in terms of angle and distance from from the origin. In that case the plots wouldn't look like soda cans. So I guess I missed Francisco's point.

Sorry Mark, the example of the soda cans was to illustrate volumes in cylindrical coordinate systems, not color spaces, maybe not the best example for this topic.

Regards

[fdisilvestro]

Francisco, there are different Delta E functions, but almost all are a square root of squares (of real numbers). So the function is always positive, hence you can never get back to point of origin.

Sandy

I remained a little confused by this explanation and looked for info about deltaE. It turns out that what you refer as the square root of squares of real numbers which is always positive is nothing more that the simple equation for calculating the "Euclidian" distance in a Cartesian, orthogonal three dimensional space such as Lab.
I understand the issues abou lab not being really perceptually uniform, the refinements made to the deltaE formula and that we cannot consider the gamut volume as an exact representation of the number of colors, but I think your explanation is not correct.


Regards

[sandymc]

I remained a little confused by this explanation and looked for info about deltaE. It turns out that what you refer as the square root of squares of real numbers which is always positive is nothing more that the simple equation for calculating the "Euclidian" distance in a Cartesian, orthogonal three dimensional space such as Lab.
I understand the issues abou lab not being really perceptually uniform, the refinements made to the deltaE formula and that we cannot consider the gamut volume as an exact representation of the number of colors, but I think your explanation is not correct.


Regards

Francisco, my answer about delta E wasn't really intended as a part of my explanation; it was just in the nature of a direct answer to the question you raised. For the purposes of the explanation that I was trying to make, whether or not the function is conservative doesn't matter. (Perhaps it should, but that's not part of what I was saying).

Sandy

[EricV]

The problems only happen when you conflate color stimuli with the subjective internal experience of perceiving color. If you don't do that, the problems go away and the answer to how many colors can AdobeRGB contain is: AdobeRGB doesn't contain colors — it is a measure of color stimulus which is continuous, not discrete.

Perfectly true, and does help clarify the discussion here, but really just kicks the problem down the road.  The fundamental question can be phrased a little more carefully -- "How many perceivably different colors can be described by AdobeRGB (or sRGB) stimulus values?"

[Jim Kasson]

I've mentioned it in passing, and it seems nitpick-y, but in context I think it's worth saying again. We are not assigning numbers to colors. We are assigning numbers to stimuli. That's why the numbers are often referred to as tristimulus values.

It's a very important distinction. We have all this light bouncing around at various wavelengths. We can measure this stimulus with tools that allow much finer precision than we can see. We can take these measurements, manipulate them with equations, then send the numbers to a device that will output some other stimulus. Colorimetry will tell us if these two stimuli — the input and the output — should match under certain conditions even when the spectral components are not the same. If we call these what they are: stimulus values or tristimulus values in the specific case, we will be in agreement with the existing literature and avoid the complications of calling them colors. Colors (as defined by subjective experience) can't be measured, color stimuli can. There's nothing complicated or philosophically problematic with any of this. The stimuli are properties of the outside world, which we can measure and reproduce with as much precision as current technology allows. This is reflected in the continuous nature of the math. The problems only happen when you conflate color stimuli with the subjective internal experience of perceiving color. If you don't do that, the problems go away and the answer to how many colors can AdobeRGB contain is: AdobeRGB doesn't contain colors — it is a measure of color stimulus which is continuous, not discrete.

Mark,

I can't argue with your facts. However, in the jargon of color management, specifications for colorimetric stimuli are called colors, to distinguish them from specifications for stimuli at the device level, which used to be referred to as colorants. When used in that sense, the word color does not refer to the subjective experience, but to a way to specify a (actually, any) spectrum that produces the encoded tristimulus values.

Reserving the work color for the (so far unquantifiable) subjective experience removes it from the realm of colorimetry, and thus, paradoxically, from the realm of color management, does it not?

But perhaps I miss your point. We have exchanged the color scientist's secret handshake -- W&S -- so I have great respect for your opinions.

Jim

[MarkM]

But perhaps I miss your point. We have exchanged the color scientist's secret handshake -- W&S -- so I have great respect for your opinions.

While I appreciate that, my copy of W&S is an indication of aspiration rather than knowledge. I'm just a photographer (and hobbyist programmer)— definitely not a color scientist.

Maybe it is just kicking the semantic can down the street, but it seems possible to remove the concept of color as defined by subjective experience from this problem entirely. The numbers in colorimetry represent measurements of the physical world. They have an aspect of psychophysics that comes from the color matching functions but these are only there to tell us if two stimuli match to an observer.

If we just stick with that, we can say that the number of colors (or device colors or colorants) in a colorspace are not countable.

You might then say, yes, but colorimetry also gives us color difference formulas, and we can use those to ask questions such as: given a set of values like the 16.7M 8-bit numbers and a known threshold, how many of those colors fall within this threshold and have a difference smaller than the average person can perceive? You can also ask in a space like ProPhotoRGB how man fall outside the spectral locus? These are easy-to-understand questions and easy to answer.

This avoids all the messiness and baggage that comes with statements like 'color is defined by what we see', or 'color by definition is a perception.' None of that is necessary to give a clear and true answer to this problem. Traditional colorimetry handles it quite well if you ask the right questions, are clear that the color difference formulas are not perfect, and understand that the answer needs to meet the same conditions of the rest of the system (i.e. similar viewing conditions, etc.)

After re-reading this it seems obvious to the point of banality, but it helped clear the problem up in my head if nothing else. I'm not sure how well I represent Andrew's proposed audience, but if he explained this to me in these or similar terms, I would understand it quickly.

[fdisilvestro]

Francisco, my answer about delta E wasn't really intended as a part of my explanation; it was just in the nature of a direct answer to the question you raised. For the purposes of the explanation that I was trying to make, whether or not the function is conservative doesn't matter. (Perhaps it should, but that's not part of what I was saying).

Sandy

Sandy,

On second thoughts it seems to me the issue that lead to the discussion is a matter of the way the terms are used (nomenclature)

The term deltaE is being used for two different things and this lead to confusion.

One use, which I think is the correct one, is the "distance" between perceivable colors.

The second use is refer to deltaE as the unit of the axis in the Lab space (at least in CIE 76). We could say that the units of those axis are one deltaE apart, but it is not technically correct to call the unit of the axis deltaE.

It would be equivalent to measure physical distances in meters and instead of calling the axis meters and volumes cubic meters, to call the axis units distances and volumes cubic distances

Naming the volume of a color gamut cubic deltaE would be like calling volumes of physical objects cubic distances.

Since the distance is deltaE, could the unit be "E"?

Regards

[digitaldog]

You might then say, yes, but colorimetry also gives us color difference formulas, and we can use those to ask questions such as: given a set of values like the 16.7M 8-bit numbers and a known threshold, how many of those colors fall within this threshold and have a difference smaller than the average person can perceive? You can also ask in a space like ProPhotoRGB how man fall outside the spectral locus? These are easy-to-understand questions and easy to answer.

I'm with you Mark. Especially after my useful conversation with Steve where is said: Color values refer to human perception and specifically to colorimetry.
- Lab, Luv, XYZ, Yxy, etc are all color values. delta-E refers to differences in color values.

Yet there appears to be some fudge factor here as colorimetry can define colors that are not perceived by humans.

In the email:
Andrew Rodney: I'm confused too because you said: - color values refer to human perception and specifically to colorimetry.
Perception is key isn't it? How can two color values that appear the same be two different values based on human perception?

SU: Well, I think you’re splitting language hairs here. Colorimetry assigns numbers to colors and those numbers are supposed to correspond to human perception. Sometimes it will assign slightly different numbers to colors that we see as the same. That’s ok because the differences between colors is not a discrete function like in the device space. Color differences vary from 0.000001 on up. We *estimate* that the average person will see colors as different when that difference is 1.0 or larger but it depends on many factors - the age of the person (corneal yellowing, etc), the fatigue level, diet, etc, etc.

Maybe I'm splitting hairs but I don't see how we can have it both ways. Or maybe we have to and as such, just stay away from using colorimetry to define numbers of Color Values.

[Alan Goldhammer]

This is one of the best threads ever on LuLa.  When this gets wrapped up (if it ever does) can the web manager publish this whole dialogue in a PDF version for future reference.  I've learned an awful lot!

Alan

[sandymc]

Sandy,

On second thoughts it seems to me the issue that lead to the discussion is a matter of the way the terms are used (nomenclature)

The term deltaE is being used for two different things and this lead to confusion.

One use, which I think is the correct one, is the "distance" between perceivable colors.

The second use is refer to deltaE as the unit of the axis in the Lab space (at least in CIE 76). We could say that the units of those axis are one deltaE apart, but it is not technically correct to call the unit of the axis deltaE.

It would be equivalent to measure physical distances in meters and instead of calling the axis meters and volumes cubic meters, to call the axis units distances and volumes cubic distances

Naming the volume of a color gamut cubic deltaE would be like calling volumes of physical objects cubic distances.

Since the distance is deltaE, could the unit be "E"?

Regards

My concern here is, maths aside, that lab itself does not actually have units in an absolute sense.

Perhaps I might draw an analogy (usually a bad idea, but anyway). Say you wanted to talk about the volume of a truck. You could do so by talking about how many grand pianos could fit in a truck. Now that is a reasonable approximation and is "relatively correct" in the sense that I think Chromix means it. So a truck that can take 4 grand pianos is bigger than one that can take 3 grand pianos - correct in a relative sense. So the measure is not exactly wrong. And maybe in some situations, to some audiences, the grand piano approximation of truck size would be useful.

But grand pianos is not actually a measure of volume.  

Sandy

[Steve Upton]

Hi all,

I thought I'd weigh into this thread as it's an interesting one and I've already got skin in the game, so to speak.

My distinction between 'colors' and 'device values' is intended as a simple distinction between two worlds. A Chinese wall of sorts. On one side are all the numbers attempting to quantify color perception in humans. On the other side are all the settings we send to devices (or get from devices).

I find this distinction helps lay people get their heads around color management issues and often use my Color of Toast story to illustrate the idea. Interestingly, I also find that it helps with trouble-shooting color management issues. Knowing the flavor of device value (working space RGB vs 'my display' RGB vs desktop printer RGB, etc) often teases apart conflated ideas and flawed workflows.

Anyway, the more subtle arguments that may occur on one side of the wall (colors vs color stimulus values, perceptible vs imperceptible colors, etc) or the other (8 bit vs 16 bit, should we call a working space a device, etc) don't detract, I think, from the overall value of the distinction.

That said, in point form to save time:

- color vs color stimuli - I appreciate the distinction but I'm not sure of it's value (especially considering the 2 sides of the wall). In my world the color that exists in solely in the head doesn't play a very big role. I call everything on this side of the wall color and it doesn't seem to cause issues or create confusion. But to each their own.

- gamut volume numbers - I believe they are calculable and have (relative) value but certainly should be taken with a large grain of salt.

- unique color values - due to flaws in Lab, differences in human perception and other factors, I think that another large grain of salt needs to be used - but again, they have value. If only to reduce computational requirements in software packages.<g> I really like Eric's description "How many perceivably different colors can be described by..."

- working spaces as virtual devices. I say why not? From a workflow or color calculation point of view why should I care if the ICC profile I'm using corresponds to a real device, an averaged set of devices, or an idealized device? The RGB values still mean the same thing, images convert the same way, etc. The best qualities of idealized devices are that we can make them perfect (gray balanced, no weird issues introduced by measurement flaws) and we can arbitrarily set their primary colors depending on our needs.

Also, I'm digging the geek level of this group. As a dog-eared Wyszecki & Stiles owner, I appreciate just how deep the rabbit hole can go.

regards,

Steve Upton

[Jim Kasson]

- working spaces as virtual devices. I say why not? From a workflow or color calculation point of view why should I care if the ICC profile I'm using corresponds to a real device, an averaged set of devices, or an idealized device? The RGB values still mean the same thing, images convert the same way, etc. The best qualities of idealized devices are that we can make them perfect (gray balanced, no weird issues introduced by measurement flaws) and we can arbitrarily set their primary colors depending on our needs.

In that case, can we agree that colorimetrically characterized virtual emissive displays -- like sRGB & Adobe RGB -- are decidedly on the "color" side of the wall, not the "device" side?

Jim