Dan Margulis Applied Color Theory

LAB Conversions and White Point

___________________________________________________________________________

   Date: Mon, 25 Oct 2004 15:47:36 +0100
   From: Andrew Haley
Subject: Photoshop Lab

Let’s say you have a file saved as Lab.  Open it in Photoshop and render it for printing.  What white point does Photoshop assume for the source (Lab) file?  I can’t find this information anywhere.

Thanks,
Andrew.
___________________________________________________________________________

   Date: Mon, 25 Oct 2004 09:43:32 -0600
   From: Andrew Rodney
Subject: Re: Photoshop Lab

on 10/25/04 8:47 AM, Andrew Haley wrote:

 Let’s say you have a file saved as Lab.  Open it in Phososhop and
 render it for printing.  What white point does Photoshop assume for
 the source (Lab) file?  I can’t find this information anywhere.
 
D65.

Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________

   Date: Wed, 27 Oct 2004 16:15:41 -0400
   From: Henry Davis
Subject: Re: Photoshop Lab

Does this mean that an actual D65 lab measurement taken of a real object whose values are then used for reference in an object in a lab document would, when passed through an accurate profile(Absolute/RelCol), render as close to that measured color as is possible?  If that color is within the gamut of the device, then shouldn’t it nail it?  As I think you get the drift of the question, would you kindly offer some amplification on this topic.

Henry Davis
___________________________________________________________________________

   Date: Wed, 27 Oct 2004 16:24:31 -0600
   From: Andrew Rodney
Subject: Re: Photoshop Lab

D65 is a mathematical description (an Illuminant.). These Illuminants are tied to the relative energy distribution that corresponds to the radiation emitted by a so called black body. Therefore D65 has a spectral energy distribution that closely matches that of a black body at 6500K. Lab too is a totally synthetic color model. Passing this through a color space conversion could produce any number of results based upon the source profile.

Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________

   Date: Wed, 27 Oct 2004 17:18:26 -0600
   From: Nathan Wade
Subject: Re: Photoshop Lab

On Oct 27, 2004, at 2:15 PM, Henry wrote:

 Does this mean that an actual D65 lab measurement taken of a real
 object whose values are then used for reference in an object in a lab
 document would, when passed through an accurate
 profile(Absolute/RelCol), render as close to that measured color as is
 possible?Ê If that color is within the gamut of the device, then
 shouldn’t it nail it?Ê As I think you get the drift of the question,
 would you kindly offer some amplification on this topic.Ê

It has the potential to render “accurately” but context is everything. Relationships between neighboring colors are probably more important than an exact colorimetric match.

And, technically there is no such thing as a D65 Lab measurement. Measurements are usually in XYZ and are converted to Lab (and sometimes require chromatic adaptation). The point is that a D65 Lab value is not the end-all-be-all description of the color of the original physical object.

Trying to nail every color colorimetrically is probably a recipe for heartbreak. If you only have a few colors you’re concerned about, it might be possible to make a device link profile that converts your special Lab colors to device-dependent colors as you dictate.

Nathan Wade
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 11:06:21 -0400
   From: Henry Davis
Subject: Re: Photoshop Lab

Exact colorimetric matching is the goal, in this case.  This isn’t about pretty pictures.  Why is it that a profile cannot be made whose job is exact coloirmetric matching, especially in a hypothetical case where every color is within the gamut of the output device?   Why is a device link the better option for exact matching?

Henry Davis
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 12:18:57 -0400
   From: Henry Davis
Subject: Re: Photoshop Lab

Then a Lab document can be expected to convert properly through an output profile, even though there is no “source” profile in the Lab document?

Henry
 
On Oct 28, 2004, at 11:58 AM, Andrew Rodney wrote:

  If you built the file in PS, then use, it’s a direct LAB creation.
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 10:21:30 -0600
   From: Andrew Rodney
Subject: Re: Photoshop Lab

on 10/28/04 10:18 AM, Henry wrote:

 Then a Lab document can be expected to convert properly through an output
 profile, even though there is no “source” profile in the Lab document?

Properly? That would depend on the definition of properly and of course how the output profile behaves. Then there are other issues that could rear it1s ugly head (which CMM, does the CMM have a bug etc). Use ACE.

Andrew Rodney
http://digitaldog.net
___________________________________________________________________________
 
   Date: Thu, 28 Oct 2004 12:44:55 -0400
   From: Henry Davis
Subject: Re: Photoshop Lab

By properly, I mean, an output profile that produces an exact colorimetric match when printed - when the target color is within the gamut of the output device.  Are profiles expected to misbehave with colorimetric conversions when the target color is in gamut?  If so, and if it’s bugs, this isn’t good to hear.  The definition of a colorimetric conversion isn’t about getting unexpected results from in-gamut colors.

Henry

On Oct 28, 2004, at 12:21 PM, Andrew Rodney wrote:

  Properly? That would depend on the definition of properly and of
 course how the output profile behaves. Then there are other issues
 that could rear it’s ugly head (which CMM, does the CMM have a bug
 etc). Use ACE.
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 12:17:03 -0600
   From: Andrew Rodney
Subject: Re: Photoshop Lab

on 10/28/04 11:20 AM, Henry wrote:

 I see, thanks.  Do you think there’s a need for a profiling product that
 allows the user to custom define the number of swatches, and their values,
 specifically for producing colorimetrically accurate profiles.  Such a product
 would have no need for perceptual tables nor interpolations for colors not
 included in the swatch set and out of gamut colors wouldn’t be an issue.
 
There are products that allow you to build custom targets with various numbers of patches. The biggest patch sample I1ve personally measured was over 10,500 patches and even with a very fast auto Spectrophotometer, that took an hour. This is far short of the 16.7 million I indicated before and the profile was 5mb in size! It was slightly better then the standard target from this package which asks for 918 samples.

One other important point. Colorimetrically correct data isn1t always preferable. First, profiles don1t know anything about images, only devices. Second, I1ve seen many cases where clients and end users prefer imagery that isn1t Colorimetrically correct compared to imagery that is.

Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 13:20:42 -0400
   From: Henry Davis
Subject: Re: Photoshop Lab

I see, thanks.  Do you think there’s a need for a profiling product that allows the user to custom define the number of swatches, and their values, specifically for producing colorimetrically accurate profiles.  Such a product would have no need for perceptual tables nor interpolations for colors not included in the swatch set and out of gamut colors wouldn’t be an issue.

Henry
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 10:52:27 -0600
   From: Andrew Rodney
Subject: Re: Photoshop Lab

Output profiles are far from perfect (colorimetrically). You1d literally have measure 16.7 million patches to build a profile that used no extrapolation in conversions. You1d end up with a profile larger then many images as well. So all kinds of 3assumptions2 are made. I1d be hard pressed to believe that a prefect colorimetrically correct profile exists but I could be wrong.

Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 14:36:44 -0400
   From: Henry Davis
Subject: Re: Photoshop Lab

Yes, printing pretty pictures isn’t the goal for this hypothetical, only the accurate reproduction of in-gamut colors.  Thanks for the input.

Henry
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 15:53:31 -0600
   From: Andrew Rodney
Subject: Re: Photoshop Lab

on 10/28/04 9:06 AM, Henry wrote:

  Why is a  device link the better option for exact matching?
 
Device links are useful for doing CMYK to CMYK conversions because there1s no trip to LAB; a link is a one way street. A device link is like two profiles joined at the hip. The source and destination are fixed and specified and the conversion is direct. That1s useful when you want to control CMYK to CMYK with control over black generation (no trip to LAB). So instead of CMYK-LAB-CMYK, you get CMYK-CMYK.

Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 19:43:00 -0400
   From: Jim Rich
Subject: Re: Photoshop Lab

Henry,

I have only seen a few reasons for colorimetric matches.  And  those situations are for printing  products that require a numerical (say LAB) product color matches  and pantone colors. And there might be more than I have mentioned,

Practically speaking,  most if not all  continuous tone images are that are reproduced are pretty pictures and that the quality of the color is in the eye of the person who pays the bill.

But that is my view of color reproduction and has been for some 30+ years.
 
As for getting more accurate color matching. Theoretically, the  way to do that is to measure all of the colors you need reproduced in to your calibration software that was created to handle all of those measurements. That could be millions of measurements. Then use  those measurements as the basis for color matching in that calibration system. Now if that type of system was m commercially available, that calibration model  would  costly and just not practical.

 ICC profiles use a relatively small sample size  of the colors in  your process. Then the profiling application uses some science,  color engineering skill  and color interpolation from the samples to try and do color matching.  In the ICC world  color management software is taking an educated guess at colors that were not measured.

Is color matching with ICC profiles perfect? Nope there are trade-offs.  But  IMHO, ICC profiles generally work quite well.
 
And some of  todays  profiling applications do a great job  and are practical and effective as a calibration strategy.

My .02.

Jim Rich
___________________________________________________________________________

   Date: Thu, 28 Oct 2004 16:27:24 -0700
   From: Mike Russell
Subject: Re: Photoshop Lab

Henry wrote:

 I see, thanks.  Do you think there’s a need for a profiling product
 that allows the user to custom define the number of swatches, and
 their values, specifically for producing colorimetrically accurate
 profiles. Such a product would have no need for perceptual tables nor
 interpolations for colors not included in the swatch set and out of
 gamut colors wouldn’t be an issue.

You’ve got my curiosity going re your application.

I’m imagining it’s something like matching a finite set of hues for a clothing catalog.

Mike Russell
www.curvemeister.com
www.geigy.2y.net
___________________________________________________________________________

   Date: Fri, 29 Oct 2004 12:03:01 -0400
   From: Henry Davis
Subject: Re: Photoshop Lab

Right.  For trade and product specific colors, it would be nice to have product that shortcuts the trial and error process of finding the correct recipe for a color (that is in-gamut).  Output profiling target files are composed of either of RGB or CMYK values rather than Lab values, but their associated reference files have Lab values, and this is where I’m a bit confused.  In order to obtain the reference file Lab values, I’m guessing that they could be made from actual measurements, or, they could be assumed values based on the Lab values from a particular RGB or CMYK model.  Earlier in this tread, the particular Lab that Photoshop uses was in question, and Andrew said that it is D65.

That there is interpolation involved in profiling and conversion is understood.  However, it seems to be almost too easy to ask why a Lab reference value of say 76, -6, -20 (which is a light blue and is in gamut for the press), can’t be located without interpolation from a custom profile of the press.  The press target sheet that, in this case has maybe 1000 blue patches, is measured and listed in Lab values.  The job is done when either the desired value is found in the list (Bingo), or the closest value found is identified.

For input profiles, I’ve made my own reference file from a Macbeth color chart (which works a tad bit better than the one supplied because of the target’s age), but the printer targets are either RGB or CMYK.  Where do those reference file values come from?

Suppose you had a target Lab value, as with an input profile used for scanners and digital cameras, but it was obtained from your fabric or plastic part.  You also have good gamut description of the press.  Is it too much to ask that a profile conversion(colorimetric), or a profiling package if need be, should provide the exact values needed to render that color?

This would be terrific for dealing with the new batch of fabrics or “gotta match” projects that are real world print challenges.  Heck, they do it at the paint store, and they use a lot more colorants than 4.  I’ll almost bet that they use some colorimetric profiling that is not intended to reproduce nice pictures, but rather accurate individual colors.

Henry Davis
___________________________________________________________________________

   Date: Fri, 29 Oct 2004 12:05:14 -0400
   From: Henry Davis
Subject: Re: Photoshop Lab

Thanks for the input.  How do they do it at the paint store?
___________________________________________________________________________

   Date: Fri, 29 Oct 2004 18:18:22 -0400
   From: Jim Rich
Subject: Re: Photoshop Lab

Henry,

It is interesting that you ask that.

The short answer is they have a paint mixing machine and  a guy with experience on that machine.

I just visited my paint store with a paint swatch and I was prepared to give them the LAB value of the color so they could put that in their computer and paint mixing system that mixes the paint.

While I was there it was clear that the paint manufacturer had built a paint mixing system that allows the guy behind the counter to measure the paint swatch and have the paint machine mix the colors based on the measurement. It all  seemed quite simple but I am sure it took years and lots of money to develop a tool that a hardware store person could work and get the color mixed properly. Kind  of like the ICC system. After you spend some time and money working with the ICC tools and dtp apps it becomes relatively easy.
 
Jim Rich
___________________________________________________________________________

   Date: Mon, 1 Nov 2004 11:00:00 -0500
   From: Henry Davis
Subject: Re: Photoshop Lab

Jim,

From this thread, I have gathered that exact colorimetric matches is not quite so in the ICC arena, yet I have had several amazingly accurate matches at different paint stores.  It didn’t appear to take a great deal of understanding or hands on science on the part of the employee, either.  The sample was measured, the formula was followed, and the color matched.

Henry Davis
___________________________________________________________________________

   Date: Mon, 1 Nov 2004 10:14:39 -0800
   From: Mike Russell
Subject: Re: Photoshop Lab

Henry wrote:

 From this thread, I have gathered that exact colorimetric matches is
 not quite so in the ICC arena, yet I have had several amazingly
 accurate matches at different paint stores.  It didn’t appear to take
 a great deal of understanding or hands on science on the part of the
 employee, either.  The sample was measured, the formula was followed,
 and the color matched.

My guess is they used curves and not profiles :-)

Mike Russell
www.curvemeister.com
www.geigy.2y.net
___________________________________________________________________________

   Date: Mon, 01 Nov 2004 19:10:51 -0500
   From: Jim Rich
Subject: Re: Photoshop Lab

Henry,

Hmmm. My posts were meant to point out that if  a company ( or an individual) were to take the time to measure all of the colors possible with a set of colorants and put them into an automated    color matching system that they would stand a chance at getting exact color matches without to much pain and suffering after the initial development.

But as I pointed out that strategy is not practical.

As for icc profiles not being as good as a theoretical system that I proposed, well.....that is a little too much extrapolation.

You can do color matching anyway you want but... if you take a look at what it takes to get an imaging system calibrated and keep it calibrated you would see that ICC profiles are not only cost effective but they do  a good if not very good job of color matching. Will the results  always be a colorimetric match? Probably not. Will it look good? Probably.

How far will the colors be away from a colorimetric match? Probably not too
far, but it wont be exact  as  you were asking for.
 
Jim Rich
___________________________________________________________________________

   Date: Tue, 2 Nov 2004 11:30:50 -0500
   From: Henry Davis
Subject: Re: Photoshop Lab

Jim,

I don’t want to be misunderstood as being overly critical of profiling.  I have read discussions about how to improve upon the current solution, and where do we go from here talk, but in all of that, all I hear is the limitations etc., etc.  In this thread, I was presenting a scenario that seemed plausible, but I have been surprised to read that many experts in this field consider the topic a technical dead end.  Your comment that it isn’t practical does sound reasonable, and at least a bit more open minded, but the paint companies certainly believed it was practical. If there is to be an evolving future for profiling and color management, I would be disappointed if exact colorimetric matching was already off of the table.

Is there an evolving future ahead for profiling, or is this as far as the technology can go?


Henry Davis
___________________________________________________________________________

   Date: Tue, 2 Nov 2004 13:44:03 -0600
   From: Jim Bean
Subject: Re: Photoshop Lab

hello jim & henry,

I may not be understanding this thread.. but certainly there is a huge difference between matching one opaque color of paint, utilizing unlimited resources and matching a large number of transparent colors produced from a limited pallet and output simultaneously on a variety of media.  jim bean
___________________________________________________________________________

   Date: Tue, 2 Nov 2004 15:21:32 -0500
   From: Henry Davis
Subject: Re: Photoshop Lab

Jim,

These are good points.  Since there are more than 4 colorants available to the paint store, I wonder if this complicates or simplifies the process of matching a particular color?  At any rate, my participation in this thread was born out of ignorance.  I thought that predicting the output values for an exact colorimetric match of an in-gamut color might be one of the future goals of profiling.

Henry Davis
___________________________________________________________________________

   Date: Wed, 3 Nov 2004 11:31:08 EST
   From: Dan Margulis
Subject: Re: Photoshop Lab

Henry writes,

These are good points.  Since there are more than 4 colorants
available to the paint store, I wonder if this complicates or
simplifies the process of matching a particular color?

It doesn’t matter. Mixing paints is an entirely different subject. Paint has no detail, no variation in darkness or color, so all you have to worry about is one type of measurement, which can be accomplished in a variety of ways.  Plus, you can’t mix up a paint that’s out of the gamut of paints that can be mixed. And you view it in a normal environment next to other things that are unrestricted as to their colors. It’s possible, therefore, to come up with a test that determines absolutely whether the paint is close enough to the desired color to declare it a “match”.

If a CMYK picture contains an item that theoretically is of a known color, no such agreement is possible. First of all, even assuming that a machine would measure it as a certain color, there’s no particular reason to think that a human would agree. For example, if there’s a reasonably bright blue, within the CMYK gamut, and we try to match it “colorimetrically”, the viewer will think it’s *too* blue, because he’s unaccustomed to really bright blues in CMYK pictures.

Second, there wouldn’t be any way to measure it because there wouldn’t be agreement on what point to measure. If a car is supposed to be PMS 357, it’s only going to be that color in areas in which we consider the lighting to be normal. Everywhere else, it’s going to be lighter or darker, even if the hue doesn’t change (which it can). Two different people are likely to pick two different points for the “normal” point to measure the color.

Third, even assuming that the basic color is within CMYK gamut, it’s fairly common for it to go out of gamut as it gets lighter. Therefore, we have a situation where half the object is in gamut and half isn’t. The only way to get a natural-looking result is to tone down the parts that *are* in gamut, deliberately not matching them colorimetrically.

At any rate, my participation in this thread was born out of ignorance.  I
thought that  predicting the output values for an exact colorimetric match of
an in-gamut color might be one of the future goals of profiling.

No. That’s one of the *past* goals. What you are speaking of is what is known as “calibrationism”, the idea that there’s a perfect conversion method that works in all cases.  When you’re trying to make a critical match, some human intervention is usually necessary in CMYK. The point of profiling should be to try to reduce the need for the intervention, not to eliminate it.

Dan Margulis
___________________________________________________________________________

   Date: Wed, 3 Nov 2004 12:21:44 -0500
   From: Henry Davis
Subject: Re: Photoshop Lab

Thanks, Dan.  Then the approach is still one of noodling until satisfied with the match.
And, if a profile correctly calculates the closest starting values, this could be a time saver.

Henry Davis
___________________________________________________________________________

Adobe Photoshop training classes are taught in the US by Sterling Ledet & Associates, Inc.