Pages: [1]   Go Down

Author Topic: Comparing I1Profiler "Smart Patches" on an Pro1000  (Read 840 times)

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2117
Comparing I1Profiler "Smart Patches" on an Pro1000
« on: October 31, 2020, 06:22:26 pm »

The following charts show the percentage distribution of deltaE's for a Canon Pro1000 with glossy using Ethan's list of regular grids with maximum near neutrals from 581 to 2945 patches. See this:

https://forum.luminous-landscape.com/index.php?topic=118987

The distributions were created using all, in gamut, integer L*a*b* values, and a subset where a* and b* were within +/- 2 of zero (aka: Near Neutrals).

The Argyll profiling software was used with the following approach. An 8.2k patch set with a 20x20x20 grid and 40x3x3 sub grid along the device RGB neutral axis to create and print 9, 957 patch charts on the Pro1000. These were then read by an i1iSis 2 XL and the RGB/LAB values were processed to create a reference ICC profile.

This reference profile was then used to create RGB/LAB value pairs using the RGB values from the I1Profiler's "smart patch generator" at the counts that produce maximum near neutral extra patches.

Also included for comparison is the XRite 1 page default iSis 957 patch set as well as the "packed" 1 page set I designed that also is 957 patches.

There are two charts. One using deltaE2000 and one using deltaE1976.

Solid lines are, in increasing Patch size (and accuracy) 581, 815, 1105, 1457, 1877, 2371, 2945. These form 7, sold lines on the two charts.

Dashed lines are both 957 patches:
Red: I1Profiler's Default iSis which has 900 patches for the main grid and 57 near neutrals.
Black: My "Packed" set with 855 main grid and 102 near neutrals.

The distribution charts show the deltaE v percent of L*a*b* printed values that are lower. Thje left side is for the full, printable gamut. The right side is the near neutrals (a* and b* are between -2 and 2).

DeltaE 2000:


DeltaE 1976

Logged

JRSmit

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 916
    • Jan R. Smit Fine Art Printing Specialist
Re: Comparing I1Profiler "Smart Patches" on an Pro1000
« Reply #1 on: November 01, 2020, 04:02:34 am »

Interesting Dough, a question popped up in my mind: would a well made argyll icc profile be better than an i1profiler profile assuming almost same no of patches including near neutrals?

The area where I in particular would like to get further improved is the bottom quarter, so from L=25 to L=0, and specifically for fine art matte papers.

« Last Edit: November 01, 2020, 07:05:24 am by JRSmit »
Logged
Fine art photography: janrsmit.com
Fine Art Printing Specialist: www.fineartprintingspecialist.nl


Jan R. Smit

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2117
Re: Comparing I1Profiler "Smart Patches" on an Pro1000
« Reply #2 on: November 02, 2020, 12:25:34 am »

Interesting Dough, a question popped up in my mind: would a well made argyll icc profile be better than an i1profiler profile assuming almost same no of patches including near neutrals?

The area where I in particular would like to get further improved is the bottom quarter, so from L=25 to L=0, and specifically for fine art matte papers.

The relative positions of the lines are pretty similar for I1Profiler created flags. At least for the ones I've checked. but I haven't looked at the full set. I1Profiler can't be used to batch process profiles which is how I make the Argyll ones and it's just painful to sit here and make them manually.

That said, there are a few differences. Argyll does much more smoothing of patch values than I1Profiler. Running a reverse (AtoB) conversion of the patch values used to create a profile produces a much higher average dE comparing the input Lab values than I1Profiler.  This isn't necessarily bad. The additional smoothing Argyll does may produce better profiles when a printer's response is lumpy. But if the printer's response is very smooth, I1Profiler may produce better profiles. In either case I haven't seen any difference visually when the images are in gamut.
Logged

Ethan Hansen

  • Full Member
  • ***
  • Offline Offline
  • Posts: 148
    • Dry Creek Photo
Re: Comparing I1Profiler "Smart Patches" on an Pro1000
« Reply #3 on: November 10, 2020, 04:39:45 pm »

Interesting Dough, a question popped up in my mind: would a well made argyll icc profile be better than an i1profiler profile assuming almost same no of patches including near neutrals?

The area where I in particular would like to get further improved is the bottom quarter, so from L=25 to L=0, and specifically for fine art matte papers.

A direct Argyll to i1Profiler isn't possible with the same target. Argyll uses an algorithm that maximizes spacing between color points while i1Profiler's main target section is a regularly spaced grid. Process an Argyll target through i1P and you get a mess. The fitting spline algorithms i1Profiler uses for the base profile appear to either require or at least work best with gridded input points. An i1Profiler target run through Argyll gives an adequate profile but not as good as one made with the same number of points generated by targen. For gridded data, the measurement resolution scales with 1/#of measurements. A pseudo random sequence such as Argyll uses converges (best case) at a rate of 1/log(#of measurements). That's a noticeable improvement. Also, with gridded points you can get unlucky if the fitting spline has nodes that match up with the data grid, giving flaky results.

You can use Argyll in two stages by first building a profile from a smaller default target then creating a second target based on the first profile for "pre-conditioning" (-c flag) and add more neutrals with the -n flag. That should tease further detail out of the mud on matte papers.

That said, there are a few differences. Argyll does much more smoothing of patch values than I1Profiler. Running a reverse (AtoB) conversion of the patch values used to create a profile produces a much higher average dE comparing the input Lab values than I1Profiler.  This isn't necessarily bad. The additional smoothing Argyll does may produce better profiles when a printer's response is lumpy. But if the printer's response is very smooth, I1Profiler may produce better profiles. In either case I haven't seen any difference visually when the images are in gamut.

Argyll does indeed perform more smoothing than i1Profiler. The difference is in the spline algorithms used. See here and Graeme's response. Whether the smoothing is good or bad depends on the measurements and the printer behavior. We find that some printers behave pathologically in certain areas. A profile can smooth out some of the variation, but care needs to be taken not to introduce visible artifacts in neighboring areas of the color space. The other case is a measurement anomaly; e.g. misprinted target, gunk on the print, or random instrument error. These points should either be ignored (our approach) or heavily smoothed. Argyll is definitely less sensitive to the noise than i1Profiler. I give the nod to i1P when making a single pass profile with an adequate number of points (>= 2371) on a poorly linearized printer.

Color geek digression: Our code does an initial fit to spectral data with an M-estimator (either Tukey's biweight or Huber's depending on data and profile type). Weights are given to the points, with measurements that are far out of line being ignored. Depending on the measurements, we go either to non-polar barycentric rational interpolation or kriging (a probabilistic model originally developed for geostatistics). We then shove the perceptual and saturation tables through an optimization run to smooth out response. On systems where we have a large historical data set, we can perform a principal components analysis to initialize the optimizations with the computed eigenvalues.
Pages: [1]   Go Up