Pages: [1] 2   Go Down

Author Topic: Comparing Argyll and I1Profiler printer profiles  (Read 8347 times)

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Comparing Argyll and I1Profiler printer profiles
« on: November 09, 2017, 12:53:46 pm »

Preliminary results using Epson 9800 with Costco Glossy and M2 (uV cut) measurements.

I made an I1Profiler profile using defaults except for selecting V2 instead of V4 and max quality for the LUTS. This creates LUTs with 37 grid points. I also made 3 Argyll profiles. The default LUTs, "high" and "ultra." These have 17, 33, and 45 grid points respectively. Note that "ultra" is strongly discouraged. It's also very slow.

Argyll has numerous options for determining Perceptual and Saturation intent tables. The default, where no Perceptual mapping is selected, is to create identical tables to Colorimetric. I have not explored the Perceptual gamut options.

All profiles were correctly generated and didn't map paper black point to L=0. That is they did not incorporate BPC which, for instance, the canned profiles that came with my 9800 did.

All Argyll profiles produced better neutral tone tracking than the I1Profiler profile. However, the I1Profiler produced the lowest overall error for the 426 Lab patches which broadly cover the printer's gamut.

Ave dE00 for the 426 Lab patches
  Argyll -qm: 0.6709
  Argyll -qh: 0.5041
  Argyll -qu: 0.5420
  I1Profiler HighQual: 0.4815

Ave dE00 for the 37 neutral patches from L=5 to 95 in steps of 2.5
  Argyll -qm: 0.6309
  Argyll -qh: 0.4653
  Argyll -qu: 0.4811
  I1Profiler HighQual: 0.6708

Much of these errors, albeit small, are due to the 9800 printer itself. The i1iSiS is an exceptionally consistent instrument that contributes negligibly to the variance but the printer, and to a lesser degree the paper are larger intrinsic sources of error. Both programs have options, the Argyll one better described, to accommodate these errors but I have just used the defaults for this comparison.

It would appear the defaults result in better use of the additional neutral patches for the Argyll program while the I1Profiler's are better matched to the overall grid spacing.

I briefly looked at how OOG colors were mapped to the gamut surface for the two programs. They are radically different. Particularly at lower L* values. However, it isn't clear which would be preferable. There are likely big differences in how things like synthetic Grainger images would print. Significant OOG issues should always be handled by softproofing in any case.

One other oddity is that the L* tracking at low levels of L* (5 to 15) is significantly better (smoother) with the I1Profiler than Argyll. Even so, the Argyll L* tracking over the entire L* range more than made up, yielding significantly better neutral tracking.

Attached are the tif file that is i1iSiS ready and the CGATS file needed to initialize the chart reader. The CGATS file is not a target file, just a filler to initialize the patch locations. For those interested the RGB values are high resolution ProPhoto and are what produces the LAB values in the tiff image. Setup should be US letter Profile using i1iSiS defaults. The tif file should be printed just as any ordinary image using Photoshop manages color using Abs. Col., and selecting the profile/printer combo to test the accuracy of color reproduction.

Edited:
I've added the i1iSiS CGAT files for the scans of the LAB image using Argyll's -qh option (33 grid points) and the I1Profiler with the quality setting at max (37 grid points).

Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #1 on: November 09, 2017, 02:15:08 pm »

Now for out of gamut differences.

Protocol: Starting from a set of Lab circles in a* and b* with constant L* and slices from L* = 5 to 95 in steps of 5.

1. Initial image is in Lab space and contains a grid with grid lines every 10 units of a* and b* and a saturation radius of 125.
2. Image is converted to device space using Colorimetric w/o BPC.
3. Device space RGB values then converted back to Lab space then to ProPhoto RGB.

Since ProPhoto contains all in gamut colors the resulting images can be used to determine what color is actually printed using Rel. Col. The two images of Lab circles from Argyle and I1P show that out of gamut colors at low luminance are mapped to more closely retain L* at the expense of saturation.

Specifically, looking at the L*=10 circle (top, second from left), and looking at the positive b* axis, b* clips at about 10 while L* increases to about 17.  OTOH, the Argyll profile extends L* to just over 30 which allows b* to increase to about 30 instead of clipping at 10. Pretty much the same deltaE 1976 for both but very different approaches.

In general the I1P seems to prefer mapping OOG colors to retain luminance at the expense of earlier clipping of a* and b*. than Argyll. Neither is clearly superior for all images. What looks better on an image with significant OOG colors depends on the image. However, this sort of thing is why soft proofing should always be done

Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #2 on: November 09, 2017, 05:42:26 pm »

And now for details of the in gamut color printing accuracy. As noted earlier Argyll makes good use of the extra neutral patches in the profile target in tracking the neutral axis for higher quality B&W while I1Profiler gets slightly better color over the boarder set of in gamut colors.

Attached are charts showing the tracking accuracy along the neutral axis for both profile products. Each chart goes from L*=5 to L*=95 in steps of 2.5. There are 4 lines on each chart. L*, a*, and b* are shown as the difference between the ideal response and the measured response. dE00 is the deltaE2000 between the ideal neutral and the measured  neutral.

The line colors are as follows:
L* err: Cyan
a* err: Orange
b* err: Yellow
dE00:   Magenta

The Argyll profile is far superior to the I1Profiler except at the endpoints of L*=5 and 7.5 and L*=95.  L*=7.5 is in gamut so the fairly large error in that one location is a bit odd. But everywhere else the Argyle profile significantly outperforms.

Also included is a histogram of 426, in-gamut color dE00. The I1Profiler has slightly less dE00 than the Argyll profile but they are reasonably close. 99% of the patches have dE00s less than 1 from both profiles.

One would be hard pressed to see any difference in prints of in-gamut images using either of these profiles. Even when viewed in a graphic arts setting with 2000 Lux.
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #3 on: November 11, 2017, 01:57:42 am »

Here's the reverse lookup histograms for the profile patch generation set. This set has 15 grid points (15^3 RGB points) plus neutrals and near neutrals for a total of 3557 patches. When scanned these produce spectral data as well as LAB values for each patch along with the RGB values that printed those LAB colors.

The A2B1 tables are colorimetric and typically provide excellent inversion. In Photoshop terms this is assigning the generated profile to an image that contains the 3557 RGB values in the patch set then converting to Lab colorspace selecting Abs. Col. The resulting pixels will have Lab values close to the measured ones in the scanned CGATS file.

The 3557 was broken down into the full set of 3557 patches and the subset of near neutral patches. Each was run through A2B1 and adjusted per ICC for Abs. Col. These were then compared to the measured Lab values from the i1iSiS scan. This was done on the Argyle profile with option -qh and on the I1Profiler profile generated with max LUTs. Both were 16 bits LUTs.

This produced 4 histograms. Two each for the full patch set and two each for the near neutral patch set.

For the full patch set the I1Profiler produced by far the most accurate mapping from A2B1. with dE00 of less than .10. The Argyll profile reverse lookup had a broader distribution with an average dE00 just under .40.

However, for the near neutral patch subset, the results were flipped but close. The Argyll profile ave dE00 remained about .4 while the I1Profiler dE00 increased to .5.

Interesting.

« Last Edit: November 11, 2017, 02:18:30 am by Doug Gray »
Logged

GWGill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 608
  • Author of ArgyllCMS & ArgyllPRO ColorMeter
    • ArgyllCMS
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #4 on: November 11, 2017, 08:27:31 pm »

If you're doing these comparisons using AbsoluteColorimetric and a non Argyll CMM, then one source of difference is that by default Argyll uses a Chromatic Adaptation for the absolute<->relative conversion of the table data, to give a better visual result for the white point relative intents people use most of the time, as well as better compatibility with widely used profiles such as sRGB and AdobeRGB. For a fuller explanation, as well as a way of telling ArgyllCMS to create profiles using the ICC "Wrong Von Kries" white point conversion, see here, and the documentation here.
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #5 on: November 11, 2017, 09:35:13 pm »

If you're doing these comparisons using AbsoluteColorimetric and a non Argyll CMM, then one source of difference is that by default Argyll uses a Chromatic Adaptation for the absolute<->relative conversion of the table data, to give a better visual result for the white point relative intents people use most of the time, as well as better compatibility with widely used profiles such as sRGB and AdobeRGB. For a fuller explanation, as well as a way of telling ArgyllCMS to create profiles using the ICC "Wrong Von Kries" white point conversion, see here, and the documentation here.
Thanks Graeme, I was actually considering the adaptation issue in selecting M2 for the measurements so that adaptation errors would be minimal on my OBA heavy paper. M2 moves the white point chromaticity (xy) quite close to D50. The issue is one reason V4 profiles have an optional separate table for Abs. One that no vendor actually implements that I know of. However, since the paper's WP is Lab(95,-1,-2) adaptation differences would be very small.

I'll have to run your program with that option to create the wrong von kries and see if it makes a different. If so no doubt it's quite small.

IIRC, the last release of V2 clarified the issue for printers (Bradford for displays but not printers) and they settled on simple scaling (wrong von kries). Reversing that should produce the correct Abs Col values. Adobe's CMM as well as Matlab does that. So your option should produce better matching with the most common CMMs. I use whatever Matlab uses (LCMS I think) and Adobe's ACE. They consistently match within hundredths of a dE.

But on a related topic, the excellent B&W response from the Argyll s/w suggests you are doing a better job incorporating the higher resolution grid data in the neutrals and near neutrals. This is particularly interesting because the I1Profiler profile uses 37 grid points while Argyll uses 33 and my 9800's native device response is rather lumpy.
Logged

aaronchan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 617
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #6 on: November 11, 2017, 11:59:06 pm »

Thanks for your great research.

May I ask how did you print your CMYK chart?
Via which RIP? Or was there any other method?

Thanks
Aaron

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #7 on: November 12, 2017, 01:02:00 am »

May I ask how did you print your CMYK chart?
Via which RIP? Or was there any other method?
Aaron

Aaron,

I'm not sure I understand what you are referring to.

The charts are in 16 bit Lab space, not CMYK, and printed on an RGB printer. Since they are in Lab space they go directly into the ICC PCS and are processed by the CMM and printer profile. I find them quite handy for viewing gamut slices from L=5 to L=95 in steps of L=5. Just applying the out of gamut mask in view proof shows the gamut limits quite nicely though one should keep in mind that the gamut mask takes effect only after the dE exceeds about 6 for printer profiles. At least in Photoshop. Any dE error > 0 shows the mask for matrix profiles.

I can post a zip file of the chart in Lab space if you would like.
Logged

GWGill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 608
  • Author of ArgyllCMS & ArgyllPRO ColorMeter
    • ArgyllCMS
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #8 on: November 12, 2017, 02:20:28 am »

IIRC, the last release of V2 clarified the issue for printers (Bradford for displays but not printers) and they settled on simple scaling (wrong von kries).
Sort of - by lumping it into the 'chad' tag for displays, simultaneously disabling Absolute Colorimetric intent for them.
Logged

aaronchan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 617
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #9 on: November 12, 2017, 12:04:06 pm »

Aaron,

I'm not sure I understand what you are referring to.

The charts are in 16 bit Lab space, not CMYK, and printed on an RGB printer. Since they are in Lab space they go directly into the ICC PCS and are processed by the CMM and printer profile. I find them quite handy for viewing gamut slices from L=5 to L=95 in steps of L=5. Just applying the out of gamut mask in view proof shows the gamut limits quite nicely though one should keep in mind that the gamut mask takes effect only after the dE exceeds about 6 for printer profiles. At least in Photoshop. Any dE error > 0 shows the mask for matrix profiles.

I can post a zip file of the chart in Lab space if you would like.

I am very sorry
Since I recently got stuck in the cmyk workflow, so by somehow, the cmyk thing just got stuck in my mind

My bad...


Sent from my iPhone using Tapatalk

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #10 on: November 12, 2017, 01:51:33 pm »

Sort of - by lumping it into the 'chad' tag for displays, simultaneously disabling Absolute Colorimetric intent for them.

Lol.  So many ways to interpret "standards."  ICC decided to (re)define Abs. Col. differently than the CIE so that even Abs. Intent adapts the WP going from sRGB to the PCS so 255,255,255 remains 255,255,255 in a D50 RGB space.  I noticed Microsoft's ICM shifts the white point, contrary to current ICC advice while Adobe's ACE does not in adherence to ICC.

That aside, Microsoft's ICM has, in places, significant conversion errors round tripping from sRGB to Adobe RGB and back. while Adobe's ACE rounds with only small errors strictly just using Rel. Col.
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #11 on: November 12, 2017, 03:54:14 pm »

If you're doing these comparisons using AbsoluteColorimetric and a non Argyll CMM, then one source of difference is that by default Argyll uses a Chromatic Adaptation for the absolute<->relative conversion of the table data, to give a better visual result for the white point relative intents people use most of the time, as well as better compatibility with widely used profiles such as sRGB and AdobeRGB. For a fuller explanation, as well as a way of telling ArgyllCMS to create profiles using the ICC "Wrong Von Kries" white point conversion, see here, and the documentation here.

I set the environmental variable so the profile is consistent with current ICC specs and it did reduce the ave dE00 when inverting (AtoB) Abs. Col. from about .44 to .38. While not a lot I wasn't expecting much given the M2 white point is so close to neutral. I then used the option -r .1 instead of the default, .5, and, as anticipated, it reduced the ave dE00 further down to .18 which is still a large amount compared to the I1Profiler (.07) but I suspect the I1Profiler is overmatching and doing no significant smoothing building the AtoB tables.

Nice work Graeme. I think I'm going to be donating a bit of change to your effort. The more I look at it the better I appreciate what you've accomplished.
Logged

Alan Goldhammer

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4344
    • A Goldhammer Photography
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #12 on: November 12, 2017, 04:06:03 pm »

Nice work Graeme. I think I'm going to be donating a bit of change to your effort. The more I look at it the better I appreciate what you've accomplished.
And you cannot beat the price of the software!!!!  (note:  I've also contributed regularly over the years to this effort).
Logged

Ethan Hansen

  • Full Member
  • ***
  • Offline Offline
  • Posts: 148
    • Dry Creek Photo
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #13 on: December 07, 2017, 07:51:48 pm »

But on a related topic, the excellent B&W response from the Argyll s/w suggests you are doing a better job incorporating the higher resolution grid data in the neutrals and near neutrals. This is particularly interesting because the I1Profiler profile uses 37 grid points while Argyll uses 33 and my 9800's native device response is rather lumpy.

B&W has always been a weakness for X-Rite and GMB in days of old. When we still used ProfileMaker, I tried feeding targets with greatly increased neutral resolution. It did improve B&W accuracy and smoothness at the cost of creating nasty artifacts elsewhere. I repeated the experiment with an older version of i1P (v. 1.6 something) and saw much the same behavior. Argyll definitely does a better job here.

Measuring mathematical accuracy only gets you so far. The problem is that photography is fundamentally aesthetic. A profile with superior overall accuracy that creates ugly transitions in a few skin tones isn't as useful as a less accurate but better behaved one.

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 20630
  • Andrew Rodney
    • http://www.digitaldog.net/
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #14 on: December 07, 2017, 08:46:25 pm »

Post optimization in i1Profiler, with a big honking set of neutrals does help. Marc Levine while at X-rite produced two such targets to add here,900 or 2502 patches of just neutral values. You don't want to go there without an auto Spectrophotometer but again, I've seen improvements in gray balance doing so.
Logged
http://www.digitaldog.net/
Author "Color Management for Photographers".

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #15 on: December 08, 2017, 11:46:36 pm »

B&W has always been a weakness for X-Rite and GMB in days of old. When we still used ProfileMaker, I tried feeding targets with greatly increased neutral resolution. It did improve B&W accuracy and smoothness at the cost of creating nasty artifacts elsewhere. I repeated the experiment with an older version of i1P (v. 1.6 something) and saw much the same behavior. Argyll definitely does a better job here.

Measuring mathematical accuracy only gets you so far. The problem is that photography is fundamentally aesthetic. A profile with superior overall accuracy that creates ugly transitions in a few skin tones isn't as useful as a less accurate but better behaved one.

I tracked down the reasons why Argyll outperformed I1Profiler in Rel Col. even with a bunch of added neutrals and near neutrals. Turns out that I1Profiler does not use pre or post adjustment curves available in ICC profiles (essentially gain curves for the individual channels). The 9800 has a peculiar gain shift around L* in the high 80s. It occurs at the same point there is a large b* shift. Even with 37 grid points the LUTs can only track is so much. Argyll profiles put in the gain/slope changes in the post curves allowing the smaller grid spacing to work more effectively.

It can be quite interesting to examine profiles with ProfilerInspector which is available at www.color.org
Logged

GWGill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 608
  • Author of ArgyllCMS & ArgyllPRO ColorMeter
    • ArgyllCMS
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #16 on: December 11, 2017, 06:39:39 am »

The 9800 has a peculiar gain shift around L* in the high 80s. It occurs at the same point there is a large b* shift.
I guess that could be a dark/light ink crossover point.
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Comparing Argyll and I1Profiler printer profiles
« Reply #17 on: December 13, 2017, 01:59:04 am »

I guess that could be a dark/light ink crossover point.

This occurs in the high 80's L* region. The paper's white point is L*95. Under magnification, the inks used are LLK, LM, LC, and some Y. It's also apparent in the AtoB1 tables of the canned 9800 profiles which are quite good for Epson paper so it appears to be an intrinsic property of the 9800.
Logged

Ernst Dinkla

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4005
ArgyllCMS + best B&W workflow ?
« Reply #18 on: January 02, 2019, 08:26:29 am »

Old thread but I like to keep that information at hand in this thread.

With the Z3200-PS I have done some ICC profile creation with the route as described by Mark Lindquist;
http://z3200.com/Making_Profiles_For_HP-Z3200_Printers.htm
I have not changed the ArgyllCMS settings and used the ICC GEN interface settings for the jobs. There are differences between the profiles made with ArgyllCMS and profiles made with both HP Color Center and iPublish. Mainly the yellow with ArgyllCMS being more pure where the other ones (both near identical) have warmer yellows. I assume that it is the difference between the Rel. Col. default of ArgyllCMS versus Perceptual settings in the other two profile creators. I did not make the iPublish one but the target was printed on my Z3200. Neutrals with ArgyllCMS were already excellent and the tone range seemed to be good too.

I will come back to the color aspects and I think that the command line settings can not be avoided for creating a better workflow in color.

But right now I am more interested in B&W printing and how I can optimize both the ArgyllCMS profile and an appropriate B&W workflow for that profile + the Z3200 in color mode + application CM. I think I do not need the B&W mode and avoid the B&W profiling like I did in the past with QTR's B&W profile creation. The UCR in the HP Z3200 driver + printer is already very strong and the greys and photoblack inks are neutral by themselves, the matte black is as warm as in other pigment inksets though. So a neutral B&W image going through this route ends already in a near quad ink lay down but for the matte black and paper white adaption that might have some color dots mixed in. Main jobs are done on matte papers like Photorag.

So far my files for B&W were based on Gamma 2.2 ;  AdobeRGB/sRGB/Grey 2.2 profile (greyscale).  In ArgyllCMS perceptual rendering in printing should require settings in ArgyllCMS for either sRGB or AdobeRGB and the profiles are dedicated then.  Changing my B&W workflow I wonder whether I can stay in the Relative Colormetric rendering for both the profile creation and my workflow and what kind of assigned color space the files then should have and how BPC then can be used. The aim is as wide a dynamic range as possible on the paper including the Dmax and the next goal, a good tone range. Matte papers so more compression will happen. If that route is unwise and I will be better off with for example a dedicated AdobeRGB CM setup for the profile creation and workflow then I am listening too. I would take out black point compensation issues but I understand perceptual can be quite arbitrary too in that part of the tone range.


Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
March 2017 update, 750+ inkjet media white spectral plots
Logged

Alan Goldhammer

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4344
    • A Goldhammer Photography
Re: ArgyllCMS + best B&W workflow ?
« Reply #19 on: January 02, 2019, 08:40:13 am »


So far my files for B&W were based on Gamma 2.2 ;  AdobeRGB/sRGB/Grey 2.2 profile (greyscale).  In ArgyllCMS perceptual rendering in printing should require settings in ArgyllCMS for either sRGB or AdobeRGB and the profiles are dedicated then.  Changing my B&W workflow I wonder whether I can stay in the Relative Colormetric rendering for both the profile creation and my workflow and what kind of assigned color space the files then should have and how BPC then can be used. The aim is as wide a dynamic range as possible on the paper including the Dmax and the next goal, a good tone range. Matte papers so more compression will happen. If that route is unwise and I will be better off with for example a dedicated AdobeRGB CM setup for the profile creation and workflow then I am listening too. I would take out black point compensation issues but I understand perceptual can be quite arbitrary too in that part of the tone range.

Hi Ernst,

When you create your ArgyllCMS profiles how many additional gray step patches do you add?  I do a step of 50 for my Epson 3880 if I am going to use the normal print driver (otherwise I use QTR with the B/W option).

Alan
Logged
Pages: [1] 2   Go Up