Pages: [1]   Go Down

Author Topic: Comparing Argyll and i1Profiler Profile accuracy.  (Read 652 times)

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2173
Comparing Argyll and i1Profiler Profile accuracy.
« on: February 25, 2022, 05:25:33 pm »

I've been using my program "i1patches" to compare profile accuracy on Canon Pro1000 with glossy paper. I'm consistently getting better accuracy with i1Profiler for both the i1Profiler's default 957 patch set for the i1isis and my own optimized patch set. I used the default settings in i1Profiler and the following for Argyll where the "-r" option (smoothness) is varied:

set ARGYLL_CREATE_WRONG_VON_KRIES_OUTPUT_CLASS_REL_WP=1
txt2ti3 -v po957_M2.txt o957
colprof -v -r .3 -qh -D o957.icm -O o957.icm o957

The program generates measurement files for the default and optimized patch sets as well as an independent set patches for testing accuracy. The independent set consists of 1250 total patches. 52 are along the device neutral, 724 patches are low saturation which correspond to colors typically found in photos, and 724 patches across the full printer gamut. The 1250 independent patch set is combined with the two profile patch sets and randomly spread across multiple printed pages which are printed on the same printer run. This eliminates variations caused by printing profile charts, creating profiles, and printing separate color verification runs. Details of i1patches are outlined here:

https://github.com/doug3236/i1Patches

Each profile was then tested against the 1250 patch set to evaluate color accuracy. i1Profiler produces significantly more accurate profiles. The optimal -r setting for Argyll's "colprof" is .03 but varies little except at the extremes. Here's the results:



Note this was copied from an Excel spreadsheet. The headers "Device Neutrals" "Full Gamut" and "Low Saturation" shouldn't be separated. They refer to the two columns below which are deltaE 1976 and Delta E2000 calculations.

Note: These patch sets were optimized for i1Profiler. Argyll has a wide assortment of algorithms for patch creation and the advantage i1Profiler has may well be specific to i1Profiler charts. I expect to look at some of the Argyll targen options for patch creation. These will then be transformed into i1Profiler compatible RGB CGATs files for scanning with the i1isis and processed with Argyll.

If anyone has any suggestions as to good printer patch creation settings with targen I would appreciate the info.
« Last Edit: February 25, 2022, 11:25:32 pm by Doug Gray »
Logged

GWGill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 596
  • Author of ArgyllCMS & ArgyllPRO ColorMeter
    • ArgyllCMS
Re: Comparing Argyll and i1Profiler Profile accuracy.
« Reply #1 on: February 26, 2022, 07:23:25 pm »

I'm not unhappy with that. All the errors are below (expected) visual threshold, and within a small dE (0.1 - 0.2) of i1Profilers results in your test.

The effect of the default smoothness (r = 0.5) in Argyll actually depends on the number of test patches, so smaller numbers of patches default to higher smoothing since I plumbed for robustness over fit.
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2173
Re: Comparing Argyll and i1Profiler Profile accuracy.
« Reply #2 on: February 26, 2022, 09:01:13 pm »

I'm not unhappy with that. All the errors are below (expected) visual threshold, and within a small dE (0.1 - 0.2) of i1Profilers results in your test.

The effect of the default smoothness (r = 0.5) in Argyll actually depends on the number of test patches, so smaller numbers of patches default to higher smoothing since I plumbed for robustness over fit.

More like .05 for dE00 which is a bit closer to a perception metric.

I also just ran a synthetic test using the -X option in i1patches and it turned out the "-i" option in targen produced the lowest error. I used a similar algorithm in creating my optimized, 957 patch set which is on a single US letter page. It works slightly better than i1Profiler's i1isis default.

I've observed in the past that i1Profiler's error stats increase when the "cubes" are even slightly irregular. Changing the RGB levels by +/-1 randomly seems to change its algorithm. Also, i1Profiler has a bug when fractional RGB values are used. Either loaded in through a CGATs file or from their patch generator. The tif files generated and print function is 8 bits but they use the fractionals when calculating. They also round differently when saving data in the profiles or in patch files after generation. Small bugs but can cause an extra .1 or more dE errors.

And all of that is still small in comparison to the changes that occur as sequential pages are printed while the print head warms up.  And this is still below the point one can see any difference in prints. I've obsessed over it because I used to make calibration charts for a camera production line and controlling reflectance was critical.

BTW, I've modified i1patches to input ".ti1" files from targen. This makes it pretty easy to use i1Profiler to create and measure charts with an i1isis and this does not require a license. Saved CGATs measurement files can then be processed easily with Argyll. I haven't checked in the change yet to github but probably will after a bit more testing.
« Last Edit: February 26, 2022, 09:07:14 pm by Doug Gray »
Logged
Pages: [1]   Go Up