i1Profiler - 7 Color Profile Contamination

Schnitzel

Well-known member
Hey everyone,

I'm creating a 7-color ICC profile using X-rite's i1Profiler (1.7.1), and an i1Pro2. The colorants are CMYK + orange, green and blue.

In some specific areas, especially deep blue colors, I'm getting very ugly transformations. For example: C87 M57 Y1 K8 O16 G7 B98.
This is obviously wrong - the profile does not need to use 7 inks to reproduce a single color, especially when these inks are opposites (C/O, M/G, Y/B).
In addition, while the predicted delta-E to this patch was around 1, I managed to get the same predicted result manually using only cyan, black and blue.

Tinkering with i1Profiler's options did not solve this issue.

Does anyone have an idea?

Thanks!
 
Hi,

Maybe the problem comes from the black generation settings you are using? Which settings have you used? UCR?

Louis
Couleur911
 
Well, using "full black" separation or heavy GCR did not change anything dramatically. I'm chalking this up to bad color math. I'll try a different profile creation software and see what happens.

Thanks
 
Hello Schnitzel,

Bad colour math indeed. I don't believe any seven colour build should contain more than four inks at a time. You certainly shouldn't get any more than one of the extended inks in the same build.

The attached image shows a slide demonstrating how each extended ink is only required in the area of gamut that lies between itself and the CMYK gamut. (credit to Mark Samworth)

I could confirm that by flicking through the Pantone extended gamut fan deck but like everyone else, I never bought one. I do have some of the HP Indigo seven colour fan decks and I've never seen more than four in that one.

If you are a member of the FTA they have a nice webinar on the subject of profile building for seven colours from April 2016.


. A Deep Dive into Expanded Gamut Color Profiles
. Presenter: Mark Samworth, Esko

Yes, they refer to product's other than Esko.

(Pssst.... does anyone know if the FTA updated their website recently?)
​​​​​​
I also found you a you tube webinar from the same guy with many of the same slides. I've not watched it all so I can't be certain it's as detailed as you'd like, but about 23 minutes in the profile generation comes in.


https://m.youtube.com/watch?v=M8Hr52fEljg

Good luck.
 

Attachments

  • Screenshot_2017-08-21-23-10-58.png
    Screenshot_2017-08-21-23-10-58.png
    232.1 KB · Views: 335
Last edited:
Thank you, Tim, I'll have a look at the webinar. I think I have Esko's color profiling tools laying around here somewhere, I'll try to use them and see if the results are better.

I could confirm that by flicking through the Pantone extended gamut fan deck but like everyone else, I never bought one.

For some strange reason, we actually bought Pantone's Extended Gamut fan deck a while ago. I don't see any swatch that requires more than 3 inks to reproduce.


Can you post you color dataset measurments so we could make it a try ;-)

Sorry, Louis, I'm not allowed to share the measurement data.

Thanks again.
 
Hey Mike,

This is the CIELab of the color I posted: L*: 11.41, a*: 24.63, b*: -55.92.

Other winners of the "Worst Transformation Competition" are:
L*: 14.05, a*: 23.04, b*: -57.47
L*: 9.16, a*: 29.24, b*: -45.36

You can see the problem concentrates in the deep blue region of the gamut. Orange and green colors show cleaner results.
 
I´ve tried your CIELab vales with two different offset profiles (CMYKOB) made with two different profilers (including i1Profiler) and I don’t get any Orange % in the converted output values.

What is your printing process/equipment you are trying to profile?

Louis.

Hey Mike,

This is the CIELab of the color I posted: L*: 11.41, a*: 24.63, b*: -55.92.

Other winners of the "Worst Transformation Competition" are:
L*: 14.05, a*: 23.04, b*: -57.47
L*: 9.16, a*: 29.24, b*: -45.36

You can see the problem concentrates in the deep blue region of the gamut. Orange and green colors show cleaner results.
 
I'm profiling one of our 7-color presses (I work at Landa). It's basically a large format inkjet press that doesn't print directly onto the substrate.
 
Hi Schnitzel,
I visited Landa web site... very nice presses ;-) What is the ink sequence on press? which profiling target have you used?

Louis
Couleur911.com
 
The ink sequence is KBCGMOY. The profiling target is randomly generated for 7-colors by i1Profiler. 1,260 patches - too little, I know, but it's just a preliminary profile. I don't believe the low patch count is the source of this problem.
 
These questions come up regularly when one works in the realm of multicolor profiling and printing. Unfortunately, accurate information is not always so easy to find. One area of frequent misunderstanding is how multicolor separations are calculated. Output profiles do not, as a rule, impose any limitation in the number of output channels but rather use as many channels as necessary to reach the lowest deltaE to the target. The simple reasoning that one shouldn't need more than 4 colors in a 7C set to match any in-gamut color is based on the assumption that we have ideal colorants (including a really dark black) that overprint perfectly. In CMYK this assumption leads to the idea that we need only two chromatic colors plus black to build any color. But we all know that in practice images frequently require all 4 colors--this is especially obvious in building dark colors--we just can't get them dark enough any other way, and if that dark color is also saturated this is even more true. It's no different in CMYK+N: If you have a very dark red, nearly black, the build may include CMYK plus red or orange, for example, as that may be the mathematically shortest route to get that color. I realize that certain systems attempt to get around this logic in order to conserve output channels, and plates, but this is usually when converting spot colors, not images. That path uses a different conversion with different algorithms.

There is also a widespread concern that printing with two many colors will lead to screen angle conflicts. This is overstated. If one is forced to use AM screening any issues can be minimized by use of sensible angle assignments: black and violet can print on top of each other--both very, very dark, and complementary colors such as cyan and orange or magenta and green can safely share the same angle. Where they do overprint it is in very dark colors where moiré or dot-on-dot effects don't cause a problem.

I have seen nothing wrong with i1Profiler's "math" in multicolor, but results can be seriously compromised by the MC profiling charts it generates: These simply do not have enough samples of medium and light colors while greatly oversampling dark overprints. The result is a smooth-looking profile that consists largely of interpolated values where real data should be present. There is scant relief in importing third-party charts, as i1P has a bug that causes the profile to have a total ink limit of 200%. That is a problem. I can't comment on Schnitzel's example as I haven't seen it and don't know what his criteria are for "ugly": Is this assessment based on measured or visual assessment or simply the appearance of unexpected colors in the separation? It may be both, but beyond specifying the ink limits and black generation there is nothing to be done about the number of channels used to convert an image unless one is willing to accept less accurate results.
 
...I can't comment on Schnitzel's example as I haven't seen it and don't know what his criteria are for "ugly": Is this assessment based on measured or visual assessment or simply the appearance of unexpected colors in the separation? It may be both, but beyond specifying the ink limits and black generation there is nothing to be done about the number of channels used to convert an image unless one is willing to accept less accurate results.

You make some interesting points there Mike. But I'd like to go back to that ugly colour. I believe it is the numbers, and not the printed appearance, that look ugly (do tell me if I'm wrong Schnitzel)

Schnitzel originally posted the CIELab of the colour L*: 11.41, a*: 24.63, b*: -55.92.

So - its quite a dark colour ("You can see the problem concentrates in the deep blue region of the gamut. Orange and green colors show cleaner results.")

His returned colour build was C87 M57 Y1 K8 O16 G7 B98.

So that is every single ink in use ... but very little black.

Louis ran the same data with a six colour profile .... "(CMYKOB) made with two different profilers (including i1Profiler) and I don’t get any Orange % in the converted output values."

So something has to be amiss for the Orange to be needed in a 7 colour build but not a 6 colour build. Especially when there is also no Green in that 6 colour build either.

Do you believe that "C87 M57 Y1 K8 O16 G7 B98" would give you a TAC of under 200% ? ... maybe something else is going on here? GCR has been ruled out already.

The definition of the BLUE colourant perhaps?
 
Last edited:
Hi Tim: I haven't looked at Schnitzel's data and have no specific comments to make about it. I would however state again as a general principle that there is no point in second-guessing the logic applied by i1Profiler in making these builds just by looking at these percentages alone. In any event the logic used is consistent and only conditioned by the specified TAC and black generation, so just because it looks strange to you doesn't necessarily mean that the system was gripped by some temporary insanity. This does not mean that the data is good or bad--it could be either.

You may have misunderstood my comment about the 200% TAC. This bug occurs when importing third-party charts. It does not happen when using X-Rite's charts. Unfortunately X-Rite's multicolor charts are deficient in their sampling and can't make an accurate profile. The net consequence is that there is no path forward (that I know of) for making good MC profiles in this application.
 
Hi Tim: I haven't looked at Schnitzel's data and have no specific comments to make about it. I would however state again as a general principle that there is no point in second-guessing the logic applied by i1Profiler in making these builds just by looking at these percentages alone. In any event the logic used is consistent and only conditioned by the specified TAC and black generation, so just because it looks strange to you doesn't necessarily mean that the system was gripped by some temporary insanity. This does not mean that the data is good or bad--it could be either.

You may have misunderstood my comment about the 200% TAC. This bug occurs when importing third-party charts. It does not happen when using X-Rite's charts. Unfortunately X-Rite's multicolor charts are deficient in their sampling and can't make an accurate profile. The net consequence is that there is no path forward (that I know of) for making good MC profiles in this application.

Hello Mike,

Thank you so much for the clarification and the advice about using i1 Profiler.
 
I need to be clear that my remarks were general ones about the way ICC profiles work, and I would need to see your output (conversion) profile and workflow (including the rendering intent) to begin to understand why a deep violet blue got converted to that particular build. No, it is not normal or expected in this particular case, but again, the presence of an unexpected color combination in the output should not be taken as an indication that the profiling "logic" is flawed. In this case I suspect an error in your measurement data. If this were a color closer to the neutral axis the presence of complimentary colors in the build wouldn't necessarily bother me. If you can see your way clear to letting me see the data I will inspect it for you and rebuild the profile.
 
Tim, Mike,

Thank you for your input. I've had a go with EFI's Color Profiler Suite. For the discussed CIELab values, I get this conversion: C98 M6 Y0 K35 O0 G0 B89.

So there.


I understand your arguments, Mike, but still it feels more logical and intuitive to me that a deep blue color should not have orange or green in it. It is not clear to me why i1Profiler is using these colorants to make the color more neutral, instead of using black which is fairly dark and neutral (L*: 5.4 a*: 0.1 b*: -1.2).
Granted, EFI's predicted delta-E is higher than i1Profiler's, but as I've written in the first post, I managed to get lower predicted delta-E manually, by using only 3 inks.

There are other downsides to using so many colorants in a single area. Even if you are using the same screen angles for 2 opposite inks, slight mis-registrations could create bad interference effects. I'm using FM screening so Moire does not bother me, but registration in fine vector elements is still a concern.

I'm waiting to try Esko's color profiling tools.

Thanks again!
 
Let's assume for a moment that your measurement data is fine and you made what looked like a more logical build in another profiling app. The question then might be why there is so little black in your build from i1Profiler. Did you perhaps accidentally alter the black generation setting in i1P? A dark blue like that should have quite a bit of black in the build, after all. If the black isn't available orange certainly might be used, as it both counteracts the blue and adds density. This is not an error but a logical response to the data input and separation settings. Check them again. I've build thousands of profiles in i1P as well as Profilemaker, EFI Color Profiling Suite (and previously Color Manager), Monaco Profiler, ColorLogic CoPrA, Onyx, Heidelberg PrintOpen--what else? If I use the same data and separation settings I get similar builds. None of these arbitrarily limits the number of output channels as has been suggested., and all will use a complimentary color if the back channel is insufficient.

On an older or poorly maintained press registration can be a big problem already with 2 colors, let alone 5. If this is the case one should avoid vector elements composed of multiple colors and set exceptions in the device link so that they aren't converted.
 
Regarding the black usage: as I've written above, I tried to change the black settings in i1Profiler, using a medium or maximum curve, and even using "Full Black Separation" (which I presume is full strength GCR), to no avail. The changes in black usage were minimal - several percentages.

Regarding your second statement, I printed another CMYKOGB target, measured using i1Profiler, exported the measurements to EFI's CPS and created a profile in both programs using similar settings, to the extent possible. In effect the only difference is the color engine creating the profile.
i1Profiler's output: C87 M53 Y0 K7 O19 G4 B97
EFI's output: C95 M26 Y0 K30 O0 G0 B89

Again, the predicted delta-E is higher in EFI's profile (2.39 vs 0.86), but I suspect EFI has some mechanism to prevent more than 4 colorants in a single transformation.
 

PressWise

A 30-day Fix for Managed Chaos

As any print professional knows, printing can be managed chaos. Software that solves multiple problems and provides measurable and monetizable value has a direct impact on the bottom-line.

“We reduced order entry costs by about 40%.” Significant savings in a shop that turns about 500 jobs a month.


Learn how…….

   
Back
Top