Shadow Pinch

[BradFunkhouser]

Often times an art reproduction image that's been captured absolute colorimetric will fit entirely inside the gamut of the desired paper, in which case I can print using absolute colorimetric and get an excellent match that pleases the artist.

But sometimes almost the entire image fits inside the paper's gamut with just a few regions of very saturated dark colors that are below the gamut of the paper.  So, most of the image is an excellent match when printed absolute, but those saturated dark colors get mapped to the closest point on the paper's gamut (a mostly upward vector when viewed in ColorThink Pro).  That vector makes those colors vibrant, but too light in relationship to the colors around them.  Maintaining the luminosity of these colors often seems to be more important than the chroma or hue.

What strategies would you guys suggest for dealing with those few regions of out of gamut saturated dark colors?

Here's what I do currently in these cases:  in Photoshop, with the image in BetaRGB working space (slightly larger than AdobeRGB but not all the way out to ProPhotoRGB), I create a solid color layer that's got a and b values the same as the paper's black point, but is a few L values darker, then I use a luminosity mask on that layer to select from that L upwards, trailing off in a gradient at 15 to 20 L above the paper's black point.  This gently shifts the image's dark colors down and in towards a point just below the paper's black point, so when they get mapped absolute to the paper's gamut they end up slightly darker and not quite as vibrant, but with their luminosity relationships better preserved.  I guess what I'm doing is like a manual black point precompensation in working space in preparation for an absolute mapping into printer space.

This works, but it's a pain.  Is there a better way?

[rasworth]

No useful suggestions, but I do have a question - what do you mean by "captured absolute colorimetric"?

Richard Southworth

[Doug Gray]

Often times an art reproduction image that's been captured absolute colorimetric will fit entirely inside the gamut of the desired paper, in which case I can print using absolute colorimetric and get an excellent match that pleases the artist.

But sometimes almost the entire image fits inside the paper's gamut with just a few regions of very saturated dark colors that are below the gamut of the paper.  So, most of the image is an excellent match when printed absolute, but those saturated dark colors get mapped to the closest point on the paper's gamut (a mostly upward vector when viewed in ColorThink Pro).  That vector makes those colors vibrant, but too light in relationship to the colors around them.  Maintaining the luminosity of these colors often seems to be more important than the chroma or hue.

What strategies would you guys suggest for dealing with those few regions of out of gamut saturated dark colors?

Here's what I do currently in these cases:  in Photoshop, with the image in BetaRGB working space (slightly larger than AdobeRGB but not all the way out to ProPhotoRGB), I create a solid color layer that's got a and b values the same as the paper's black point, but is a few L values darker, then I use a luminosity mask on that layer to select from that L upwards, trailing off in a gradient at 15 to 20 L above the paper's black point.  This gently shifts the image's dark colors down and in towards a point just below the paper's black point, so when they get mapped absolute to the paper's gamut they end up slightly darker and not quite as vibrant, but with their luminosity relationships better preserved.  I guess what I'm doing is like a manual black point precompensation in working space in preparation for an absolute mapping into printer space.

This works, but it's a pain.  Is there a better way?

Some of my work is also repro and yes, it's not at all uncommon to encounter low luminosity, highly saturated colors in some artwork. That said the biggest issue I have using AbsCol is the shift in color at the black point when the lowest luminosity point of the paper differs from the lowest luminosity of neutral. Then, there is a significant, abrupt, shift as L drops even slightly below what can be rendered. This is quite similar to what you are encountering. It's easy to deal with using curves in my case but your approach looks like it would work as well and also deals with OOG , low luminosity colors.

Were I to need to address a problem like you have I would use a Matlab script but probably approach it very similarly to what you do in Photoshop. It would also be quite easy to make a program in C++ to automate this using Little CMS and tiflib with just a few command line parameters. I haven't run across anyone that has done this but it would be useful for these situations.

[Doug Gray]

No useful suggestions, but I do have a question - what do you mean by "captured absolute colorimetric"?

Richard Southworth

I'm sure he means "scene referred," possibly adapted to D50 white point if D50 wasn't the illuminant, which is the norm for precision reproduction. It's a kind of absolute in that the colors are rendered colorimetrically rather than the normal output referenced which typically compresses with S curves.

[BradFunkhouser]

what do you mean by "captured absolute colorimetric"?

Richard Southworth

I mean this...  File comes out of the RAW camera workflow, gets assigned the camera ICC profile that I built specifically for that copy lighting setup and that RAW workflow using ArgyllCMS, then the file gets converted via absolute colorimetric rendering intent into BetaRGB working space.  This keeps the colors as "measurably" accurate as possible in relation to a theoretical perfect white diffuser versus being scaled to make the pure white in the image file match some physical target white patch (which can never hit the true white of a perfect diffuser).   Keeps the colors "absolute"...

Maybe someone else can explain it better?  Help.  That's my understanding.  But I do know from working on hundreds of art reproduction images that it "absolutely" works, since the prints are excellent matches to the original as long as the colors in the art are inside the gamut of the printer/paper, and as long as the viewing light isn't too far away from D50.

[rasworth]

Sorry, still confused, how did the image come out of the RAW camera workflow without being rendered into some colorspace?  Or are we not talking about ACR or Lightroom?

Richard Southworth

[Doug Gray]

I mean this...  File comes out of the RAW camera workflow, gets assigned the camera ICC profile that I built specifically for that copy lighting setup and that RAW workflow using ArgyllCMS, then the file gets converted via absolute colorimetric rendering intent into BetaRGB working space.  This keeps the colors as "measurably" accurate as possible in relation to a theoretical perfect white diffuser versus being scaled to make the pure white in the image file match some physical target white patch (which can never hit the true white of a perfect diffuser).   Keeps the colors "absolute"...

Maybe someone else can explain it better?  Help.  That's my understanding.  But I do know from working on hundreds of art reproduction images that it "absolutely" works, since the prints are excellent matches to the original as long as the colors in the art are inside the gamut of the printer/paper, and as long as the viewing light isn't too far away from D50.

Google using (site:color.org "scene referred") for discussions of this subset of color processing. Yours is a pretty typical repro workflow. Abs Col is the usual intent used for printing. Whether Abs or Rel are used on the input conversion depends on how the profile is constructed but in either case the result should produce a white point that matches the illuminated scene on a perfect white diffusing surface.

[BradFunkhouser]

Sorry, still confused, how did the image come out of the RAW camera workflow without being rendered into some colorspace?  Or are we not talking about ACR or Lightroom?

Richard Southworth

It comes out of the Canon RAW processor rendered into "WideGamutRGB" but that gets changed immediately by assigning the custom built Argyll camera profile to it then converting absolute colorimetric to BetaRGB

[BradFunkhouser]

Were I to need to address a problem like you have I would use a Matlab script but probably approach it very similarly to what you do in Photoshop. It would also be quite easy to make a program in C++ to automate this using Little CMS and tiflib with just a few command line parameters. I haven't run across anyone that has done this but it would be useful for these situations.

Interesting.  A very different approach.  Thanks.

[BradFunkhouser]

Google using (site:color.org "scene referred") for discussions of this subset of color processing. Yours is a pretty typical repro workflow. Abs Col is the usual intent used for printing. Whether Abs or Rel are used on the input conversion depends on how the profile is constructed but in either case the result should produce a white point that matches the illuminated scene on a perfect white diffusing surface.

Yes.  "Scene Referred" is it.   I'd just always thought of it as "absolute"

Thanks.

[GWGill]

So, most of the image is an excellent match when printed absolute, but those saturated dark colors get mapped to the closest point on the paper's gamut (a mostly upward vector when viewed in ColorThink Pro).  That vector makes those colors vibrant, but too light in relationship to the colors around them.  Maintaining the luminosity of these colors often seems to be more important than the chroma or hue.

It would be worth you looking at how ArgyllCMS V1.9 handles that when it gets released. I made some changes in the gamut clipping code that may help a bit in regards to the behavior you describe. The changes also make the clipping much more tunable, so another approach would be to introduce an absolute colorimetric intent variant with more constant-L clipping behavior.

[samueljohnchia]

It would be worth you looking at how ArgyllCMS V1.9 handles that when it gets released. I made some changes in the gamut clipping code that may help a bit in regards to the behavior you describe. The changes also make the clipping much more tunable, so another approach would be to introduce an absolute colorimetric intent variant with more constant-L clipping behavior.

Hi Graeme, that is delightful news, thank you. I look forward to V1.9 with great interest. I am most curious to see what new changes have been made.

Regards,
Sam

[digitaldog]

Google using (site:color.org "scene referred") for discussions of this subset of color processing. Yours is a pretty typical repro workflow. Abs Col is the usual intent used for printing. Whether Abs or Rel are used on the input conversion depends on how the profile is constructed but in either case the result should produce a white point that matches the illuminated scene on a perfect white diffusing surface.

Or read this article I co-authored with Jack Holm for the ICC with the photographer in mind.
http://www.color.org/ICC_white_paper_20_Digital_photography_color_management_basics.pdf

[BradFunkhouser]

Or read this article I co-authored with Jack Holm for the ICC with the photographer in mind.
http://www.color.org/ICC_white_paper_20_Digital_photography_color_management_basics.pdf

That's a nice white paper.  I actually read that paper quite a few years ago and it helped me better understand which direction I needed to go for art reproduction.  Thanks for the link.

[BradFunkhouser]

The changes also make the clipping much more tunable, so another approach would be to introduce an absolute colorimetric intent variant with more constant-L clipping behavior.

Yes!    That could handle the topside problem too, where saturated light colors just above the paper's gamut get vectored down by absolute, ending up too dark in relationship to the in gamut colors around them.  It could also handle the times when working space white gets introduced into the image (like a white border, or white text on a colored border) and what is intended to be pure white ends up too dark because it gets vectored down onto the sloping edge of the paper's gamut instead of hitting the paper's actual white point (unless the paper's white point happens to be EXACTLY on the D50 neutral axis, which has never been the case for me). 

On some art images, I currently have to do that manual Shadow Pinch shifting I described above and also a manual White Capping (the topside version of the same pre absolute shifting method).  They are both a total pain, but worth it versus the current alternatives.  The possibility of a new alternative is exciting.  Thanks.

[Doug Gray]

Yes!    That could handle the topside problem too, where saturated light colors just above the paper's gamut get vectored down by absolute, ending up too dark in relationship to the in gamut colors around them.  It could also handle the times when working space white gets introduced into the image (like a white border, or white text on a colored border) and what is intended to be pure white ends up too dark because it gets vectored down onto the sloping edge of the paper's gamut instead of hitting the paper's actual white point (unless the paper's white point happens to be EXACTLY on the D50 neutral axis, which has never been the case for me). 

On some art images, I currently have to do that manual Shadow Pinch shifting I described above and also a manual White Capping (the topside version of the same pre absolute shifting method).  They are both a total pain, but worth it versus the current alternatives.  The possibility of a new alternative is exciting.  Thanks.

Interesting.  I've not (yet) encountered a reproduction work where the subject's white point exceeded the print paper. At least on high gamut gloss types. Even reproduction of art which has a white border on the original  should never go to L=100 since the measured white (with a spectro) should match the image Lab values. For those rare cases where artwork L could get close to 100 one can use PTFE papers. Pricey but the L value can be upwards of 99. You might wish to look into those specialty synthetic papers.

[BradFunkhouser]

Interesting.  I've not (yet) encountered a reproduction work where the subject's white point exceeded the print paper. At least on high gamut gloss types. Even reproduction of art which has a white border on the original  should never go to L=100 since the measured white (with a spectro) should match the image Lab values. For those rare cases where artwork L could get close to 100 one can use PTFE papers. Pricey but the L value can be upwards of 99. You might wish to look into those specialty synthetic papers.

I agree that nothing on the art itself generally has a much higher L value than paper white, and when it does it's not usually by enough to cause a problem.  The white I was talking about was more along the lines of white edging with a colored mat or white text on a colored mat being introduced by doing layouts in Photoshop or InDesign that include the image, but also include other elements... for portfolio book prints, art cards, art show placards, etc.   Then printing absolute becomes more problematic.  You have to deal with the white mapping in some way.

From my experience, the easiest way is to punt on absolute colorimetric print accuracy, scale the white in the image up and the black down, do the layout without worrying about absolute, and just print relative colorimetric with black point compensation.  Not as good a match for the art image, but often good enough.  And this lets you exchange the layout with other people. 

The harder way is to export the layout pages and bring them into Photoshop to do a pre absolute white shifting method in working space, what I call White Capping.  That's a pain, but then you DO get absolute colorimetric matching on the print for the art image itself, which can be the most important thing.

[Doug Gray]

I agree that nothing on the art itself generally has a much higher L value than paper white, and when it does it's not usually by enough to cause a problem.  The white I was talking about was more along the lines of white edging with a colored mat or white text on a colored mat being introduced by doing layouts in Photoshop or InDesign that include the image, but also include other elements... for portfolio book prints, art cards, art show placards, etc.   Then printing absolute becomes more problematic.  You have to deal with the white mapping in some way.

From my experience, the easiest way is to punt on absolute colorimetric print accuracy, scale the white in the image up and the black down, do the layout without worrying about absolute, and just print relative colorimetric with black point compensation.  Not as good a match for the art image, but often good enough.  And this lets you exchange the layout with other people. 

The harder way is to export the layout pages and bring them into Photoshop to do a pre absolute white shifting method in working space, what I call White Capping.  That's a pain, but then you DO get absolute colorimetric matching on the print for the art image itself, which can be the most important thing.

Relative works great when you aren't comparing side by side since white point adaption works quite well.

Sounds like you have a good handle on this little subset of imaging science. Belated welcome to LuLa!

[unesco]

It would be worth you looking at how ArgyllCMS V1.9 handles that when it gets released. I made some changes in the gamut clipping code that may help a bit in regards to the behavior you describe. The changes also make the clipping much more tunable, so another approach would be to introduce an absolute colorimetric intent variant with more constant-L clipping behavior.

Any idea you can share when v1.9 is going to appear?
I wait for that mostly hoping to get rid of pink tint mostly in shadows with when making perceptual printer profile with ColorMunki on v1.8.3.

[GWGill]

Any idea you can share when v1.9 is going to appear?

I'm hoping to start the release process shortly. As well as building and packaging for the various platforms,  I need to check that every instrument works on every platform, which takes some time, and of course if there is a problem, the whole thing needs re-doing.