Why Do We Need Colour Working Spaces

[Manfred M]

All space conversions in Photoshop are Relative Colorimetric for one thing. You only get Perceptual, Saturation, or Absolute if the printer does it.

As for the spaces everything was 8 bit then, and it's just barely adequate for Adobe RGB.

I'm not sure which version of Photoshop you are using; but I get a choice of all four rendering intents; relative, perceptual, saturation and absolute for colour profile conversions with Photoshop CC 2014. I only use perceptual and relative colormetric for all my photo work. The main difference between the two is how they handle out of gamut colours.

I have the same four options when I print, and these are there whether I let Photoshop or the printer manage the colours on my Epson 3880. I use Photoshop to do the colour management when I print in colour and let the printer do it when I print B&W

[Mick Sang]

[QUOTE=Simon Garrett;516149] Thank you for this reply. I do not pretend to understand all the concepts involved here and I freely admit to having a lot to learn. Over the years, I have studied various books and papers etc on colour management all of which usually begin with the assumption that colour working spaces are reqired. But, the reason for this is never provided. That is what I seek.

In my photography, I, too, use ProPhoto RGB for my working space. So, we are in complete agreement on that. I had hoped that Adobe Wide Gamut RGB would be a better choice, as it does not include non-existent colour. But, it’s gamut turned out to fall short of some print gamuts. So, for now at least, I am sticking with ProPhoto which is also the default space for Adobe LightRoom.

When I said "translating from it [human spectral locus] to RGB or some other space and back again" I was actually referring to CIE LAB which is the so called PCS (connection space) between profile rendering and proofing in colour management.

My question may be born of a misunderstanding of the difference between CIE LAB and the Spectrum Locus, if indeed there is one. So, I suppose the crux of the question is what is the difference between CIE LAB and the Spectrum Locus i.e. the gamut of colours visible to humans? Is the spectrum Locus not defined by the colour coordinates specified in CIE LAB? If CIE LAB and the Spectrum Locus are one in the same space where LAB is the mathematical map of colour coordinates within the Spectrum Locus, then my initial question stands unchanged. Why can we not use this as the only reference system?

If, on the other hand, CIE LAB and the Spectrum Locus are two entirely different things, then I ask why we do not use only CIE LAB as the one and only working space? Is it for some convenience only?

RGB has gamuts depending upon the device in question. The colours within those gamuts have numerical assignments. The same is true of CIE LAB and perhaps the spectrum locus of that I am unsure, as I mentioned. So, why do we subset the working spaces?

In another Forum, I sought a clear answer from those who are apparently in the know. However, I received no reply at all for quite some time. Then, after my asking again, an administrator made an attempt to no avail. So, I do appreciate your replies which I’m hopeful will finally lead me to an answer.

[Mick Sang]

Let's look at the history of a few well known colour spaces....

Thank you for the history synopsis. While I have been aware of all that, I'm quite sure others are not and it is likely very helpful to many. What I ask is why. Why did all this happen? Why do we not simply use CIE LAB for the one and only working space?

[Simon Garrett]

So, for now at least, I am sticking with ProPhoto which is also the default space for Adobe LightRoom.

It's not just the default, it's the only working space used by Lightroom. There's no choice.

When I said "translating from it [human spectral locus] to RGB or some other space and back again" I was actually referring to CIE LAB which is the so called PCS (connection space) between profile rendering and proofing in colour management.

ICC profiles are defined in terms of transforms between the profile colour space and a PCS. In practice all colour space mappings using icc profiles are done as a two-stage process via the PCS, which is either CIELAB or CIEXYZ. That's just the way it's defined.

My question may be born of a misunderstanding of the difference between CIE LAB and the Spectrum Locus, if indeed there is one. So, I suppose the crux of the question is what is the difference between CIE LAB and the Spectrum Locus i.e. the gamut of colours visible to humans? Is the spectrum Locus not defined by the colour coordinates specified in CIE LAB? If CIE LAB and the Spectrum Locus are one in the same space where LAB is the mathematical map of colour coordinates within the Spectrum Locus, then my initial question stands unchanged. Why can we not use this as the only reference system?

If, on the other hand, CIE LAB and the Spectrum Locus are two entirely different things, then I ask why we do not use only CIE LAB as the one and only working space? Is it for some convenience only?

I've not come across the term spectrum locus (or spectral locus) meaning "the gamut of colours visible to humans". In my experience the term refers to the spectral edge of the CIE diagram, not the entire area of it, nor the bottom "magenta line", which does not correspond to spectral colours. That is, it refers the colours corresponding to single wavelengths of light. See for example https://en.wikipedia.org/wiki/CIE_19...ticity_diagram.

I don't know of a "spectrum locus" colour space. The commonly-used colour spaces that encompass the entire human-perceptable colour gamut are the CIE colour spaces.

RGB has gamuts depending upon the device in question. The colours within those gamuts have numerical assignments. The same is true of CIE LAB and perhaps the spectrum locus of that I am unsure, as I mentioned. So, why do we subset the working spaces?

A working space shouldn't be a subset; it should be a superset. That is, it should be a colour space with a gamut at least as wide, and preferably wider, than any other colour space you want to use. I guess you could use CIE LAB or CIE XYZ as working spaces. However most photographers think in terms of RGB values. You can use LAB representation in Photoshop, but AFAIK that doesn't alter the working space - only the numerical co-ordinate system used within the colour space.

[xpatUSA]

Mick, in your mind, you should start separating things like CIELAB and RGB from things like sRGB and ProPhoto. The former are color models; the latter are color spaces which have gamuts and neutral points each different.

It's "Spectral Locus", by the way, not "Spectrum Locus", pedantic I know.

You said "RGB has gamuts depending upon the device in question. The colours within those gamuts have numerical assignments."

But that did not say enough. Firstly, in the RGB color model, a set of RGB numbers does not define a particular color. By that I mean that RGB = 20,126,245 in ProPhoto (Kodak ROMM) is not the same color as RGB = 20,126,245 in sRGB.

There is a well-respected calculator where we can play with numbers like that:

http://www.brucelindbloom.com/index....alculator.html

There we can find that the ProPhoto actual color above is x,y = 0.072, 0.194. The sRGB color is x,y = 0.179. 0.160. To our eyes those colors are very different! Even worse, when the ProPhoto numbers are transformed into to the sRGB space, the resulting converted color is wildly out-of-gamut in sRGB - RGB = -179,158, 228. Can't have negative reds in the RGB model, so that red would be clipped to zero and the converted color would be badly wrong.

Main point I'm making is that RGB numbers do not define a color, ever.

On the other hand, CIELAB is derived from CIEXYZ and, therefore, CIELAB numbers represent unique, actual colors and are device-independent. For example, you can specify a paint color with Hunter Labs original L*ab system. With RGB numbers per se, you can not.

Not a clear answer, I know, more of a response to what you said.

[xpatUSA]

What I ask is why. Why did all this happen? Why do we not simply use CIE LAB for the one and only working space?

I'll try and follow my previous post which I know did not directly answer this question.

We can, of course "work" in any space we like. Since CIELAB is a Profile Connection Space it could indeed be used for any intended output profile. If your brain could handle those funny looking colors a* and b* on your screen and the brightness in L* (which is not the same as the brightness Y' that we are accustomed to), then I suppose we could edit such an image. However, what we would not know is how the image would look if posted on-line or printed out at home.

In a similar vein, I used to select ProPhoto as my working space and edit happily away until the day I realized that "saving as" a sRGB JPEG often resulted in color-gamut-clipping which was not present while editing.

I'm sure that, at this point, we are talking at cross-purposes.

[xpatUSA]

I'm not sure which version of Photoshop you are using; but I get a choice of all four rendering intents; relative, perceptual, saturation and absolute for colour profile conversions with Photoshop CC 2014. I only use perceptual and relative colormetric for all my photo work. The main difference between the two is how they handle out of gamut colours.

I think Richard is referring to where, in PhotoShop, selecting 'perceptual' will not actually get you that unless the profile you're using has the full look-up tables (CLUTs) - as are typical with printer profiles. If you save for the web or for on-screen viewing, you'll get the standard ICC sRGB profile which does not have the tables, just a simple conversion matrix. For such images, PhotoShop applies 'relative colorimetric' whether you like it or not. I hasten to add that that is what I've read, not personal experience.

For use with RawTherapee, I do have a 'display class' profile with CLUTs and can successfully use 'perceptual' intent, mainly for flower shots.

[Manfred M]

Mick - I think I finally understand where you are coming from. I think the first step is what Ted has covered off; the difference between a colour model and a colour space.

Let me try this with an analogy. Let's look at something related to colour; i.e. blackbody radiation. We have a model that says that at absolute zero, i.e. there is no motion and we have a temperature of 0K. We start heating things up and get to something the size of a star and measure the core temperature, you are looking at millions of K.

You now go to the folks that manufacture devices to measure temperature and ask why they don't simply have a single temperature measurement device that can measure anything from a black hole, which is pretty darn close to absolute zero to an UV star that is quite hot. The answer is rather obviously, that such a device would be extremely expensive and impractical as you would have to measure everything from the basic subatomic vibrations of atoms at close to absolute zero to a thermometer we measure ambient temperature at, cook food at, refine or heat treat metals to the temperatures of stars.

Instead of a "one size fits all" solution, device we have a host of different instruments to measure temperature. Each one will be specialized and useful for the range it was designed for. Some will use properties of metals as they change temperature, others could rely of expanding fluids while others will measure the colour of the light emitted.

The same analogy goes for colour spaces. Each of the prevalent ones was designed for a very specific subset of the total visible spectrum, because it makes sense to do so. Coming up with a one size fits all model would likely be unwieldy and expensive to develop, so we have these colour spaces that handle very specific (oiften device dependent) needs.

[Simon Garrett]

The same analogy goes for colour spaces. Each of the prevalent ones was designed for a very specific subset of the total visible spectrum, because it makes sense to do so. Coming up with a one size fits all model would likely be unwieldy and expensive to develop, so we have these colour spaces that handle very specific (oiften device dependent) needs.

I would probably not say "subset of the total visible spectrum" but subset of the total human gamut.

The spectrum of light contains only a tiny subset of the gamut of human colours. Most colours that we can see are not spectral - a single wavelength of light - but are combinations of at least two - and generally at least three wavelengths. For example, you won't find brown, magenta or white in the spectrum - but they are all colours so far as our eyes are concerned.

The spectral colours are those around the outside of the curve in the figure below. The blue figures around the edge are the wavelength of light (in nanometres) corresponding to that spectral colour. As you can see, the only common colour space that can show any spectral colours is ProPhoto RGB. Adobe RGB and sRGB don't reach the edge of the CIE diagram at any point, so don't include any spectral colours.

The difference between the common RGB colour spaces is (roughly) in terms of how saturated are the colours they can represent. sRGB can handle moderately saturated colours, Adobe RGB includes all of sRGB plus more highly saturated cyans, greens and yellows. ProPhoto RGB includes more highly saturated colours of all hues than the other two, and includes pretty much all colours one is likely to find in nature (and more than pretty much any camera can capture, any printer can print or any monitor can display).

[Cantab]

These colour space discussions are always so much fun to read!

[Mick Sang]

It's not just the default, it's the only working space used by Lightroom. There's no choice.

That is quite true during input and processing. But, Adobe RGB and sRGB are options for export in LightRoom.

ICC profiles are defined in terms of transforms between the profile colour space and a PCS. In practice all colour space mappings using icc profiles are done as a two-stage process via the PCS, which is either CIELAB or CIEXYZ. That's just the way it's defined.

This is also usually true, unless one chooses to use a device link in which case the transforms are not done

I've not come across the term spectrum locus (or spectral locus) meaning "the gamut of colours visible to humans". In my experience the term refers to the spectral edge of the CIE diagram, not the entire area of it, nor the bottom "magenta line", which does not correspond to spectral colours. That is, it refers the colours corresponding to single wavelengths of light. See for example https://en.wikipedia.org/wiki/CIE_19...ticity_diagram.

Spectrum Locus is referred to as the Gamut of Human vision in a program called ColorThink by Chromix and is referred to as such by Mr. Andrew Rodney, a well known American colour management specialist, in the video at this link: http://digitaldog.net/files/ColorGamut.mov

A working space shouldn't be a subset; it should be a superset. That is, it should be a colour space with a gamut at least as wide, and preferably wider, than any other colour space you want to use. I guess you could use CIE LAB or CIE XYZ as working spaces. However most photographers think in terms of RGB values. You can use LAB representation in Photoshop, but AFAIK that doesn't alter the working space - only the numerical co-ordinate system used within the colour space.

I appreciate that. But, are working spaces not subsets of the so called Spectrum Locus Gamut of human vision?

[Mick Sang]

Mick, in your mind, you should start separating things like CIELAB and RGB from things like sRGB and ProPhoto. The former are color models; the latter are color spaces which have gamuts and neutral points each different.

This is helpful. Thank you. Colour MODEL vs Colour SPACE. This I must explore.

It's "Spectral Locus", by the way, not "Spectrum Locus", pedantic I know.

In a program called ColorThink by Chromix, it is Spectrum Locus. It is also referred to as such by Mr. Andrew Rodney, a well known American colour management specialist in the video at this link: http://digitaldog.net/files/ColorGamut.mov

You said "RGB has gamuts depending upon the device in question. The colours within those gamuts have numerical assignments."

But that did not say enough. Firstly, in the RGB color model, a set of RGB numbers does not define a particular color. By that I mean that RGB = 20,126,245 in ProPhoto (Kodak ROMM) is not the same color as RGB = 20,126,245 in sRGB.

There is a well-respected calculator where we can play with numbers like that:

http://www.brucelindbloom.com/index....alculator.html

There we can find that the ProPhoto actual color above is x,y = 0.072, 0.194. The sRGB color is x,y = 0.179. 0.160. To our eyes those colors are very different! Even worse, when the ProPhoto numbers are transformed into to the sRGB space, the resulting converted color is wildly out-of-gamut in sRGB - RGB = -179,158, 228. Can't have negative reds in the RGB model, so that red would be clipped to zero and the converted color would be badly wrong.

Main point I'm making is that RGB numbers do not define a color, ever.

I agree with this and this is also discussed by Andrew Rodney in the above mentioned video.

On the other hand, CIELAB is derived from CIEXYZ and, therefore, CIELAB numbers represent unique, actual colors and are device-independent. For example, you can specify a paint color with Hunter Labs original L*ab system. With RGB numbers per se, you can not.

Not a clear answer, I know, more of a response to what you said.

Yes, but I believe that is because RGB numbers are actually coordinates in reference to a given working space which act like maps. Also, this only brings to my question to mind once more. If we were able to use one master working space such as the Spectrum Locus or gamut of human vision, we would avoid the potential for colour errors based solely upon duplicate numerical values occurring in multiple working spaces.

In any case, you have given me good food for thought in your post. This, together with the other very helpful posts from those who have kindly offered their thoughts here may finally lead me to a clear answer to my question and a more complete understanding of this subject. I am grateful. Thank you.

[Simon Garrett]

That is quite true during input and processing. But, Adobe RGB and sRGB are options for export in LightRoom.

Agreed, but this discussion is all about working spaces, which is what is used "during input and processing". As I say, Lightroom always uses ProPhoto RGB colour space (but with linear TRC) as its working space.

This is also usually true, unless one chooses to use a device link in which case the transforms are not done

As I understand, in this case the two profile transforms are precalculated into a single transform. As I think of it, it's bit like saying that you have two transforms: the first to multiply by 3, the second to divide by 4. So you combine them into one transform: to multiply by 3/4.

Spectrum Locus is referred to as the Gamut of Human vision in a program called ColorThink by Chromix and is referred to as such by Mr. Andrew Rodney, a well known American colour management specialist, in the video at this link: http://digitaldog.net/files/ColorGamut.mov

Thanks for that link. I'd not come across the term used that way, nor did I find that when Googling. I've just searched the Chromix web site, and Google can't find the term there. I'll have a look at Andrew Rodney's video.

I appreciate that. But, are working spaces not subsets of the so called Spectrum Locus Gamut of human vision?

Yes, but you need some numerical frame of reference for representing the gamut of colours or a subset. What numerical framework do you suggest?

There are frameworks that represent the entire gamut such as CIE XYZ or CIE LAB, or frameworks that represent a subset, such as ProPhoto RGB, Adobe RGB or sRGB.

What else could you use?

[Simon Garrett]

Spectrum Locus is referred to as the Gamut of Human vision in a program called ColorThink by Chromix and is referred to as such by Mr. Andrew Rodney, a well known American colour management specialist, in the video at this link: http://digitaldog.net/files/ColorGamut.mov

I've just scanned Andrew's video - thanks again for the link - and I see where he refers to the gamut of human-perceptable colours as the "spectrum locus". I don't want to argue with Andrew - he has written what is probably the best book on Colour Management for photographers - but I think that term is rather misleading.

Colour is not the same as the spectrum of light. The colours of the rainbow - what we regard as the spectrum of visible light - are only those colours that sit on the curved edge of the CIE diagram. Colours away from the curved edge, including colours along the bottom "magenta line" are not spectrum colours.

Colour, in the sense of a phenomenon that humans can perceive, is the sensation we perceive as a result of the combination of the stimuli to the three sets of cone cells in our eyes - the co-called L, M and S cells. We think of them as red-sensitive, green-sensitive and blue-sensitive cells, but this is a very rough approximation. See for example https://en.wikipedia.org/wiki/Color_...lor_perception

These three sets of cells each respond to light within bands of wavelengths that correspond to different subsets of the visible spectrum. Any combination of wavelengths that creates the same combination of stimuli from the L, M and S cells is the same colour. Not "looks like the same colour", but "is the same colour", by definition.

Magenta is an example of a non-spectral colour. It can be created only be the combination of at least two wavelengths of light: one somewhere in the range 420 to 480 nanometres approximately, and one in the range 600 to 650 nanometres. No single pair of wavelengths uniquely defines "magenta" - many different combinations of red and blue will be perceived identically to us as magenta.

For this reason, colour is only indirectly related to the spectrum, and - with the greatest respect for Andrew Rodney - I think that referring to the gamut of human perceptible colour as the "spectrum locus" is rather misleading.

[Manfred M]

Yes, but I believe that is because RGB numbers are actually coordinates in reference to a given working space which act like maps. Also, this only brings to my question to mind once more.

This in itself is problematic when it comes to using these practically. Try using a wide colour space like ProPhoto with anything less than 16-bit length integers, the spacing between individual shades becomes very large and you start getting artifacts when editing (in fact you already have this issue when working with 8-bit sRGB files). One could go to a floating point system, but then colour accuracy could be influenced by rounding errors.

I don't think it necessarily solves any real world issues as one continues to have the issues around mapping device specific out of gamut rendering issues, and the issues that this creates; look at how Perceptual and Relative Colormetric impact "true" colours as they re-work out of gamut colour conversions.

[xpatUSA]

For this reason, colour is only indirectly related to the spectrum, and - with the greatest respect for Andrew Rodney - I think that referring to the gamut of human perceptible colour as the "spectrum locus" is rather misleading.

Hear, hear!

Where "Spectrum Locus" is used, rightly or wrongly, it is simply bad grammar - also known as American English . "Spectrum Locus" is two nouns used together, whereas such a phrase demands an adjective then a noun, e.g "brown bear".

The word "spectral" is an adjective - thus "spectral locus" is grammatically correct.

[rant]Sadly, we have gone from an era when I got my face slapped by the English Master for using bad English to an age when nobody gives a pooh and everything is shortened or abused to the points of being unintelligible or ambiguous.[/rant]

The hyphen is dead - long live ambiguity

[Simon Garrett]

Sadly, we have gone from an era when I got my face slapped by the English Master for using bad English to an age when nobody gives a pooh and everything is shortened or abused to the points of being unintelligible or ambiguous.

I spent recent years in a large corporate trying to cope with documents drenched in meaningless cliche and gobbledygook. They were often written as Powerpoint presentations - bullet points rather than grammatically complete sentences - leading to confusion and lack of management clarity. I could think of a few people who should be slapped in the face by your English Master. It wouldn't help, but it would make me feel better.

[Krawuntzel]

Hear, hear!

Where "Spectrum Locus" is used, rightly or wrongly, it is simply bad grammar - also known as American English . "Spectrum Locus" is two nouns used together, whereas such a phrase demands an adjective then a noun, e.g "brown bear".

The word "spectral" is an adjective - thus "spectral locus" is grammatically correct.

[rant]Sadly, we have gone from an era when I got my face slapped by the English Master for using bad English to an age when nobody gives a pooh and everything is shortened or abused to the points of being unintelligible or ambiguous.[/rant]

The hyphen is dead - long live ambiguity

I cannot withhold the (unwarranted sarcasm): OMG
Erwin

[Mick Sang]

Hear, hear!

Where "Spectrum Locus" is used, rightly or wrongly, it is simply bad grammar - also known as American English . "Spectrum Locus" is two nouns used together, whereas such a phrase demands an adjective then a noun, e.g "brown bear".

The word "spectral" is an adjective - thus "spectral locus" is grammatically correct.

[rant]Sadly, we have gone from an era when I got my face slapped by the English Master for using bad English to an age when nobody gives a pooh and everything is shortened or abused to the points of being unintelligible or ambiguous.[/rant]

The hyphen is dead - long live ambiguity

While I am still studying all of these posts and the subject matter at the core of my question, I had to stop here and agree completely with your statements regarding the misuse or abuse of the English language which are exemplified by such nail screeching phrases as "a airplane," for example where the indefinite article "an" has been lost. To each his own, I suppose. ;-)

[Mick Sang]

I want to thank all of those who participated in this discussion for taking the time & trouble to help me (and others, I'm sure) to better understand this topic. Your responses have provided enough detailed information to help me to reach an answer to my question. I am honoured to be a new member of the Cambridge in Colour community.

Best Regards to all.

Mick