A curiosity: strange conversion results from custom icc profile

[curiosity]

Converting a Lab value (7, -128, 63) into CMYK.
When I use anything but my custom CMYK profiles, I get predictable results. A very dark green; basically black.
However, if I use my profiles, I get green, minus 3rd color and minus black. Cyan and Yellow only!
So obviously it's my profile. Can anyone help with this?

I'd like to attach one of my profiles for anyone's inspection, but can't seem to do it. Invalid file type.

 

[arossetti]

Can you upload it to a dropbox or google drive and share it out?

 

[curiosity]

https://drive.google.com/file/d/0B-DGbe_ePgdJVFE2dnlVaUFMMkE/view?usp=sharing

 

[arossetti]

Yeah you have something wrong with your shadow colors/black channel. Compare your profile that has a flat plateau to the SWOP profile that has a defined tip the shadow/black. What device is this profile for and what software are you using to create the profile.

 

[curiosity]

I don't see that in the colorsync utility, unfortunately. (in fact, what I do see looks "pointier" than SWOP)
PrintOpen (4.0)
Offset sheetfed 4c press

 

[arossetti]

When the profile was made, what were the primary color targets?

 

[curiosity]

I'm sorry, but I don't know what you mean. Primary color targets?

 

[arossetti]

What was the cieLAB values you targeted for CMYK when you created the characterization chart that made the profile? Of if you don't know that what was the standard ink density for CMYK.

 

[curiosity]

Oh, that.
55, -37, -50 @ 1.4
48, 74, -3 @ 1.4
89, -5, 93 @ 1.05
16, 0, 0 @ 1.7

I get pretty close at those SIDs

Cardstock, C2S, ~93 bright, standard inks

 

[arossetti]

I'm not exactly sure and hopefully someone else can chime in however - to me the profile is bad and you should recreate it. Make sure you average multiple press sheets from different parts of the sheet.

 

[Stephen Marsh]

I'm not exactly sure and hopefully someone else can chime in however - to me the profile is bad and you should recreate it. Make sure you average multiple press sheets from different parts of the sheet.

Agreed 100%, even more so on averaging out readings from say beginning/middle/end of run.

If curiosity can provide a copy of the CGATs measurement data, then other interested parties could see what happens with their profiling software.


Stephen Marsh

 

[ddonevski]

Averaging measurements is always recommended since it reduces measurement noise. This reflects in better "round-trip" results agreement. The round trip for this profile, as reported by SampleICC Profile Dump utility does not seem too bad. SampleICC is a free library you can download from http://www.color.org/sampleicc.xalter

I also used Matlab and here's what i got:

[7 -128 63] -> B2A1 table -> [0.9961 0 1.0000 0] (CMYK displayed in 0 - 1.00 range)

[0.9961 0 1.0000 0] -> A2B1 table -> [51.7632 -71.0078 26.7930]

[51.7632 -71.0078 26.7930] -> B2A1 table -> [0.9821 0.0041 0.9620 0]

From what I see here, the behavior of this profile has less to do with measurements, and more to do with gamut mapping. When the profile was calculated, the non-existing (out of gamut) [7 -128 63] L*a*b* value was mapped to [51.7632 -71.0078 26.7930] L*a*b* value which does exist (inside or at the gamut boundary) for this process.

I've never inspected other profiles you mentioned, but what happened with this profile does not surprise me much for M channel. I wouldn't expect to find all three primaries (CMY) at the gamut boundary since it is gamut boundary... It is normally determined by primary and secondary colors, while tertiary colors fill the inside volume.
What is strange, however, is that such dark color L*=7 was mapped to such relatively light color L*=51.76. I'd expect it to get mapped to some darker color which uses some K.
I can't explain fully what's going on here, but I'm convinced it has to with how the profile creation software performs gamut mapping.

With simplified gamut visualization, your case would look something like in the below images.

 

[Jason]

Change your rendering intent to Perceptual and you'll produce a color with these values: C93.1 M40.9 Y100 K50.2. That's probably closer to what you were expecting. Perceptual works much better than Relative Colormetric for out-of-gamut colors. Your profile overall looks good, it's slightly larger than GRACoL.

 

[arossetti]

Perceptual works much better than Relative Colormetric for out-of-gamut colors. Your profile overall looks good, it's slightly larger than GRACoL.

Define works better? This is kind of a blanket statement that could be dangerous without context.

 

[curiosity]

Thank you everyone. Much appreciated.
For the record, I always average a minimum of 4 sheets.
And I've experimented with several profiles created from years past (from same program, using same parameters) and got the same results.
I'll experiment with new parameters.
If I get different results I will post them.

 

[ddonevski]

As Jason said, using Perceptual intent makes sense if your input contains such extreme out of gamut colors.
If it's important for You that Relative intent also provides the result which you expect, maybe you have a choice of different gamut mapping algorithms when creating profile. As you can see from the picture I posted, the one employed in your profile seems to preserve hue and/or chroma at the expense of lightness preservation. Some other algorithm may perform better. They usually operate in LCH space with objectives such as lighness, chroma or hue preservation.

 

[Schnitzel]

Perceptual works much better than Relative Colormetric for out-of-gamut colors.

I agree with arossetti, this statement is problematic when used categorically: since the ICC does not define how to implement the rendering intent tables, it is up to the vendor and its "secret sauce", so the results might be better for one vendor but worse for another. Also, it depends on the image contents and the viewer.

 

[Jason]

As you know, rendering intents are the rules of how to handle out-of-gamut colors with converting from an ICC profile with a large color gamut to one with a smaller color gamut. Typically this would be a RGB to CMYK conversion.

Perceptual compresses the entire gamut to fit inside the smaller color gamut. It attempts to preserve the color relationships, however it can sacrifice the gray's as a result. Generally, it works better when converting from a large color gamut to a small color gamut.

Relative Colorimetric preserves any color that is in-gamut in the smaller profile. Only the colors out-of-gamut are moved to the closest color available. It maintains better gray's however, you can see clipping in the outer edges of the color gamut where one color moves, but another does not. Generally it works better when converting profiles of similar size, typically CMYK to CMYK.

The rendering intent is defined in the color management settings of the RIP.

The ICC profile only defines a default rendering intent to be used if none were selected.

Choosing a rendering intent is a personal choice. There are pros/cons to each and it depends upon the image. Knowing how it works allows you to make educated decision about how to control images that move through your color workflow.

The issue with the OP's custom ICC profile is most likely due to the software that created the ICC profile or the settings that were used to create the profile. The tags tell me that you used the Heidelberg ICC profile creation software, but very little about the specific settings.

The GRACoL ICC profile was created using different ICC profile creation software. I'm not sure what they used, but it probably wasn't Heidelberg's.

I would suspect if the characterization data was used to create a new ICC profile using different ICC profile software such as X-rite i1 Profiler, the result would be a gamut mapping that produces the result you desire when using Relative Colorimetric.

 

[curiosity]

I have recreated more profiles that yielded same results.
Attached are the parameters I can adjust.
This program has always done well for me, but I'm at a loss as to why it's removing the contaminant colors (M & K).
Maybe it just doesn't do well with conversions from Lab, which is definitely strange, but rarely do we convert from Lab. So....IDK.

Again, thanks to all. This little exercise is keeping me busy today. A lull in the workflow has allowed me to delve further.

 

[ddonevski]

As I said in earlier post, for Colorimetric intent, out-of-gamut colors are mapped to colors at the gamut boundary. All colors at the boundary are primary or secondary (+K for darker ones). That's why M=0. You won't find tertiary colors (C, M and Y all greater than zero) at the boundary.
I'm not sure whether the choice of gamut mapping algorithm (GMA, you have it in options) will affect Colorimetric intent (or only Perceptual), but it's worth a shot. If it doesn't then I'd repeat what Jason said, the difference is in the profile creation software. Two apps apply different GMAs when calculating Colorimetric intent.

Just to correct my statement. There are some tertiary colors at the boundary, but only a limited set. Gamut boundary colors are those for which one of the channels equals either 0 or 1, and others vary in continuous [0, 1] range. Setting one channel to 0 or 1 restricts colors to faces (boundaries) of CMY cube or CMYK hypercube. See image.