Color management myths and misinformation video

[Jim Kasson]

Andrew, I've been thinking more about the video, and here's my advice.

The basic video starts with an explanation of the Lab color space, which will be necessary to understanding the gamut plots in that space. Say nothing about where it comes from. Ignore the Cartesian aspect, and treat it like a cylindrical space, without ever using the term. Explain the lightness axis, hue angle, and chroma, giving them a consistent set of accurate but not too scary names. Then plot values for an sRGB monitor in that space. Animations would be good here. At a minimum plot the whole gamut and change teh viewing angle pointing out various features: the point at the bottom, the point at the top, the places where each of the primaries reaches its maximum value while the others are zero to form the three little tweaks at the upper part of the gamut. If you can do it, show animations of each  primary running through its range, alone and in combination with the other primaries. Explain additive color along the way. Explain gamma if you want, but I think it's unnecessary and possibly confusing.

Then, with a wireframe of sRGB present for reference, show the results for an Adobe RGB display: just the gamut if that's all you have, and animations if you can do them. Point out that the sRGB gamut lies within the aRGB one.

Then explain what a working space is, and how it can be a real or idealized monitor space. Then show PPRGB's gamut. This is a little tricky, since PPRGB has a different white point than either of the other two spaces, and white point translation is beyond teh scope of the video.

Then show a couple of printer spaces, and point out how they differ from monitor spaces: more chroma available at lower L*, less at higher L* except for yellows, more dark cyans available, etc.

Plot both sRGB and some inkjet space, and show where the printer's gamut exceeds sRGB, and vice versa.

Pick an image in a biggish space, and show the gamut of colors in it.  Show what colors can't the printed, and what ones can't be displayed in sRGB.

I think you can take it from there. The difference between changing color spaces and just changing profiles, for example. No chromaticity diagrams. No discussion of linearity. No information on what goes on under the covers when you change color spaces.

If people are interested, you could do a companion video: introduction to the eye, cone responses, the color matching experiment, XYZ, xy, the ellipses, u'v', monitor gamuts in u'v', opponent color (to prepare them for Lab), Weber's law and nonlinear luminance response, Cartesian Lab, cylindrical Lab, color differences, and an optical illusion of two to show that it's not all about point color.

Jim

[digitaldog]

Jim, I was thinking of starting with the results and working backwards. That is, show the two prints I made using the Gamut Print Test image from ProPhoto RGB and sRGB output to my 3880. Hook them early, then explain WHY. I know not all the subtitles of the two prints will necessarily appear that well over the internet. I'd like the viewer to get an idea what to look for and suggest they do their own testing with the file I provide. Of course HOW to make the print tests correctly. From there, I'd explain many of the points you and other's point out about Lab and so on. When discussing how to convert the data from ProPhoto RGB to sRGB (and even Adobe RGB if they desire), I could show and explain the Assign Profile command and illustrate how that works and how colors 'go dull' when mistreated.

I'm hoping if I scan or shoot the two prints side by side and show them up close, the first take home will be, there IS a difference in the two working spaces all things being equal. I'm not only seeing vastly increased color saturation between the two, I see areas in sRGB on-print much like Bills where color detail in sRGB is mush. I'm seeing areas of detail on the fabric which looks much sharpener due to the better differentiation from the wider gamut colors. I'm seeing blues going somewhat magenta from sRGB where they look as I desire from ProPhoto RGB. Bill Atkinson's 14 balls is a real eye opener! Even if I end up having to do more than one video, I'm thinking that showing how the rubber hits the road first will hook some of the viewers into sticking around for the 'whys' of this improvement in output.

Of course their mileage will vary if they use the proper methodology to output the file using the two working spaces. So it's important they learn to use either Convert to Profile for sending to a lab or the Print dialog in Photoshop going out to their own printer. They need to ideally test both Perceptual and RelCol for the set of tests.

Of course I'm still running this around in my head. But I'm thinking for most audiences, if they are interested in the practical advantages of a wide gamut working space over a narrow one, from raw, show them the output first. Then explain how to do the testing on their own, then explain what's happening.

[Jim Kasson]

Jim, I was thinking of starting with the results and working backwards.

Sounds good, Andrew.

Jim

[Slobodan Blagojevic]

...I was thinking of starting with the results and working backwards...

+2

[fdisilvestro]

Or you can have the same hue and saturation, it just won't be as bright, which is another way of saying a larger gamut buys you volume on every dimension, not just saturation.

I understand what you mean, but I would like to specify (sorry for being pedantic)

- in a cylindrical coordinate system, such as Lch, the difference between gamuts is in one dimension only or "c" (chroma). E.g you have two cans of soda of the same height but different diameters

- in an orthogonal coordinate system such as Lab, the change is in two dimensions, "a" and "b" (the base of the parallelepiped) e.g you have two cartons of milk the same height but different bases

- in the RGB coordinate system, the difference is in the three dimensions. This is more difficult to visualize in a meaningful way to color

BTW, the use of "light cyan", "light magenta" and so on, allows high end printers to extend the gamut volume beyond standard CYMK, especially in the light colors.

Regards

[fdisilvestro]

Francisco, I found this a while ago, but your post reminded me. Thought you would enjoy it in case you get tired of the Lindbloom image: http://allrgb.com

Thanks !!

[Jim Kasson]

I understand what you mean, but I would like to specify (sorry for being pedantic)

- in a cylindrical coordinate system, such as Lch, the difference between gamuts is in one dimension only or "c" (chroma). E.g you have two cans of soda of the same height but different diameters

- in an orthogonal coordinate system such as Lab, the change is in two dimensions, "a" and "b" (the base of the parallelepiped) e.g you have two cartons of milk the same height but different bases

- in the RGB coordinate system, the difference is in the three dimensions. This is more difficult to visualize in a meaningful way to color

Francisco, a cylindrical coordinate system is a three-dimensional system. In Lab's case, the longitudinal axis is L*, the azimuth is hue angle, and the radius is chroma. A three-dimensional surface representing the gamut of a device or working space can be specified by enumerating points which define the surface, just as in RGB or the Cartesian form of Lab, whose axes are  L*, a*, and b*.

http://en.wikipedia.org/wiki/Cylindrical_coordinate_system

Or perhaps you mean something else by "dimension", in which case, please enlighten me.

Thanks,

Jim

[TonyW]

... That is, show the two prints I made using the Gamut Print Test image from ProPhoto RGB and sRGB output to my 3880.
... I'd like the viewer to get an idea what to look for and suggest they do their own testing with the file I provide.
... I could show and explain the Assign Profile command and illustrate how that works and how colors 'go dull' when mistreated.
... I'm hoping if I scan or shoot the two prints side by side and show them up close, the first take home will be, there IS a difference in the two working spaces all things being equal.
... I'm thinking that showing how the rubber hits the road first will hook some of the viewers into sticking around for the 'whys' of this improvement in output.
... Of course I'm still running this around in my head. But I'm thinking for most audiences, if they are interested in the practical advantages of a wide gamut working space over a narrow one, from raw, show them the output first. Then explain how to do the testing on their own, then explain what's happening.

Andrew, FWIW I have been following the thread with interest, although have to admit that some of the deeper discussion made my head spin what you are proposing sounds really promising to dispel some of the myths surrounding this complex subject.  The real world examples illustrating the difference will I am sure stimulate interest and I look forward to seeing your work

[fdisilvestro]

Francisco, a cylindrical coordinate system is a three-dimensional system. In Lab's case, the longitudinal axis is L*, the azimuth is hue angle, and the radius is chroma. A three-dimensional surface representing the gamut of a device or working space can be specified by enumerating points which define the surface, just as in RGB or the Cartesian form of Lab, whose axes are  L*, a*, and b*.

http://en.wikipedia.org/wiki/Cylindrical_coordinate_system

Or perhaps you mean something else by "dimension", in which case, please enlighten me.

Thanks,

Jim

Jim,

I might be using the wrong terms, but I think we are on the same page. I should have said axis instead of dimensions
Having said that, unless I'm mistaken, Lab is not a cylindrical coordinate system but a Cartesian one. The cylindrical coordinate is Lch which corresponds to L*, chroma and hue

What I tried to explain in my previous post, is that depending on which coordinate system you use, cylindrical or Cartesian, the difference between a larger gamut such as Prophoto RGB and a smaller one such as sRGB is in one (c in Lch) or two dimensions axis (a & b Lab) meaning that L does not vary

Regards

[Jim Kasson]

I might be using the wrong terms, but I think we are on the same page. I should have said axis instead of dimensions
Having said that, unless I'm mistaken, Lab is not a cylindrical coordinate system but a Cartesian one. The cylindrical coordinate is Lch which corresponds to L*, chroma and hue

You are correct. I was being sloppy and consider them both forms of Lab.

What I tried to explain in my previous post, is that depending on which coordinate system you use, cylindrical or Cartesian, the difference between a larger gamut such as Prophoto RGB and a smaller one such as sRGB is in one (c in Lch) or two dimensions axis (a & b Lab) meaning that L does not vary.

Yes, I believe we are in violent agreement.

Another way of making your point might be to say: "Chroma may be calculated using one dimension (or axis) of Lch, two of Lab, and all three of RGB."  Is that right?

Jim

[fdisilvestro]

Another way of making your point might be to say: "Chroma may be calculated using one dimension (or axis) of Lch, two of Lab, and all three of RGB."  Is that right?

Jim

Absolutely!

Thanks

[MarkM]

- in a cylindrical coordinate system, such as Lch, the difference between gamuts is in one dimension only or "c" (chroma). E.g you have two cans of soda of the same height but different diameters

A little off topic, but I'm curious about this. I've never seen a gamut plot in this coordinate space. It's clear the space would be perfectly cylindrical, but do the gamuts remain cylindrical in the space as well? I have seen gamuts plotted in HSB, but that's just a device-dependent transformation of RGB — not so useful for comparing gamuts. Do you have an example of a color space plotted in Lch? Is the L the same calculation as L* in LAB?  

[MHMG]

Or you can have the same hue and saturation, it just won't be as bright, which is another way of saying a larger gamut buys you volume on every dimension, not just saturation.

But I think I now understand the point your making — correct me if I'm wrong: while a larger space extends the upper limit of saturation it doesn't extend the limit on the L* axis. To put another way if I have an sRGB color with 100% saturation, proPhotoRGB might allow a color that is essentially 110% saturation in sRGB, but neither space can exceed 100% L*. I don't think anyone can disagree with that.

My point was that at a particular chromaticity, larger spaces allow higher lightness for that particular chromaticity.

Yes, and if truth be told there are indeed practical consequences to all of this discussion. As a good example, take Andrew's Gamut_Test_File_Flat.tiff file. Convert it from proPHotoRGB to sRGB. If you are using a wide gamut monitor you will see a big hit on color saturation and some hue shifting as well because it's a pretty extreme case of saturated colors. But more to the point, you now have a colorful sRGB image, such as it is, with many areas of color pushed right out to the the sRGB gamut edge. Now in PS try to recover some of the original scene vividness with the hue/sat tool and the adjustment layer set to "saturation" rather than normal. This will constrain the hue and lightness from shifting while attempting to amplify the chroma in this RGB color mode. You will see some movement in some of the colors, but there's really no "head room" for more chroma for many of the areas in this already colorful sRGB image, so the exercise will be rather humbling!  Next, throw that hue/sat layer out, convert the image back from sRGB into proPhotoRGB or into aRGB.  Now apply a hue/sat move using the adjustment layer set to "saturation" again, and you should find that a lot more vivid colors are "recoverable".  Certainly not in a perfectly accurate way, but definitely a much more satisfying attempt to breathe more color saturation back into the image than if you'd tried to keep editing the image in sRGB.

I think all this means the "take home" advice on choice of sRGB, aRGB, or proPhotoRGB, particularly for folks who shoot RAW, really does become simple. Stay in proPhotoRGB and 16 bit as long as you can right up until you decide what the final destination color space for the image really is. You might be giving up a little mathematical precision in the color value encoding, but this situation pales in comparison to not having pragmatic flexibility for more colorful image edits down the line.  The only deviation I'd make from that advice is when I receive an image already in a lesser color space and the colorfulness is already where I want it or I have a RAW image that will "fit" with plenty of head room into aRGB or sRGB for additional color tweaking (e.g., when I'm converting to B&W right from the start). But again, to "keep it simple" a very defensible piece of workflow advice is simply to use RAW capture, edit in ProphotoRGB until you're ready to commit the image to it's final destination color space, e.g sRGB for the web or a printer profile for a specific printer/ink/media combination, and use soft proofing wisely in order to guide your final image edits which will address the contrast, hue, and color saturation hits that will take place upon the final color conversion to the destination space.



cheers,
Mark
http://www.aardenburg-imaging.com

[digitaldog]

Going back to ColorThink, I got this useful email from Steve Upton. He said it was OK to post it. I'm still working out the details in my brain!

Hi Andrew,

There are many other finer points in your discussion but I thought I’d respond to this last one and see if that clears any of them up or inflames them or….

A long time ago I started to clearly differentiate between color values and device values - as you said above it’s semantics but it helps me be clear about the differences.

So, in that terminology, a few elementary points:

- 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.

- spectral values are not technically color values but can obviously be converted into color values so I often treat them as synonymous even though technically they aren’t

- device values are *not* color values and only become color if sent to a device or come from the device. They can be correlated with color values using an ICC profile or in a measurement file of some sort.

So…. (again, using the device vs. color semantics)

- AdobeRGB and sRGB have the same number of device values (in either 8 or 16 bit) but *not* the same number of color values. AdobeRGB can hold more *unique* color values (more below). Another saying of thinking about it is to say that if you graphed sRGB and aRGB in 3D RGB space, they would be the same size. But when graphed in 3D color space, aRGB is larger.

- gamut volumes of printing devices also have the same number of device values but different color volumes

- gamut volumes in ColorThink are calculated using color values and refer to the number of unique color values (again, more below)

- unique color values in ColorThink refer to the number of distinct colors (as per human vision) that exist in a color gamut or a color image or color list or whatever.

- In color gamuts we just use cubic Lab values, the idea being that one delta-E76 value away from a color *should* result in a perceptually different color (more on *that* below)

- in color images/lists we roughly round color values when counting unique colors to make two very similar colors (less than 1 dE) become the *same* color before counting the number of unique colors. Note: this is *not* the same thing as unique Lab values, which also has value for other purposes. The process is rough, for speed and doesn’t use delta-E but you get the gist.

- the idea behind these calculations was to provide rough approximations to allow for the comparison of color gamuts, image conversions, etc as well as to reduce the number of duplicate (or near-duplicate) color points when graphing image colors in 3D.

- it could certainly be argued that cubic Lab values and delta-E76 calculations do not represent unique color values. Only if Lab were perceptually uniform as it was intended would this be true. Delta-E 2000 and other differencing equations are attempts to correct for this non-uniformity and might lead us to better ways of calculating unique color values (but they might not, as well)

- the problems with using dE2000 equations for such purposes is that there is no color space defined by (or corrected by) dE2000, it is computationally expensive (29 separate equations for each color difference), and people still don’t think it’s the best representation of color perception - only better.

- with this in mind I decided to use Lab and dE76 for ColorThink’s initial foray into these types of calculations. Then at least the math would be reasonable and we could all agree on how it might be done, even if we also all agreed that Lab was not the best space for such things and we should pursue better things going forward.

So, if you look at the unique color values that ColorThink gives for your images in sRGB vs aRGB it makes sense that there are fewer in sRGB. ColorThink likely coalesces some sRGB values together as they are bunched tighter together.

Does this make sense? At the very least it’s been what I’ve been thinking and developing toward over the years.

Feel free to quote me on the LL list if you want.

regards,

Steve

[Jim Kasson]

A little off topic, but I'm curious about this. I've never seen a gamut plot in this coordinate space. It's clear the space would be perfectly cylindrical, but do the gamuts remain cylindrical in the space as well? I have seen gamuts plotted in HSB, but that's just a device-dependent transformation of RGB — not so useful for comparing gamuts. Do you have an example of a color space plotted in Lch? Is the L the same calculation as L* in LAB?  

A gamut plotted in Lch will look exactly the same as the same gamut plotted in Lab. The underlying triplets describing each point will be different. Yes, the L* axis is the same.

Jim

[Jim Kasson]

Going back to ColorThink, I got this useful email from Steve Upton. He said it was OK to post it. I'm still working out the details in my brain!

Andrew, thanks for this. I believe that when Steve says "device values", he means the same thing as I do when I say "colorants". A web search seems to indicate that my usage is not used very often any more. Maybe it's time to bring it back, or maybe we should use Steve's term. The word "colorant" seems to need fewer contextual clues to make sense. Perhaps "colorant values" would be a useful way of indicating that we're talking about the bits and not the atoms.

Jim

[MarkM]

A gamut plotted in Lch will look exactly the same as the same gamut plotted in Lab. The underlying triplets describing each point will be different. Yes, the L* axis is the same.

Jim

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.

[Jim Kasson]

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.?

Like the formulae on page 168 of the second edition? Yep.

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.

Yes, see my restatement of his point above: "Chroma may be calculated using one dimension (or axis) of Lch, two of Lab, and all three of RGB." Francisco agreed that that captured the essence of what he was saying.

Jim

[MarkM]

A long time ago I started to clearly differentiate between color values and device values - as you said above it’s semantics but it helps me be clear about the differences.

Yes, I think the idea of clearing up the semantics by differentiating between these two ideas with different terms came up earlier in this thread (or the last one - it's turning to mush in my head). I think it's a smart idea.

[MarkM]

Like the formulae on page 168 of the second edition? Yep.

Precisely.