i1Profiler smoothness control effect?

[rasworth]

I built color and b&w profiles for my Canon Pro-100, using Doug Gray's patch generators and following his recommendations.  The results were good, and I thought I now had conquered the profiling world.

My photographer neighbor has a Canon Pro-9000 Mkii, and armed with all of my expertise I created a black and white profile for one of his papers, Red River Palo Duro Satin - the results were kindly stated inferior, a definite green tint thru the mid-tones and shadows.  We spent hours looking at his setup, printing targets, etc., with no luck.  Finally I remembered one of Doug's posts having to do with increasing the smoothness control when using targets consisting mostly of neutrals/near neutrals and few colors.  That did the trick, with smoothness set to 80 the results were satisfactory.  The attachment is a comparison of the before (left) and after (right) profiles neutral curves, the only difference is the smoothness control increased from 50 to 80.  It's clear that upping the smoothness took down the green channel and raised the red.

We went back and used a profile I had done earlier for his printer and the same paper, for full color.  Upon closer examination we noticed it also printed slightly green in the shadows, although not as bad as my initial attempt on b&w.  Again re-profiling with the smoothness set to 80 much improved the neutral response, although not sure yet what it did to color accuracy.  The 9000 mkii has a single black, where my Pro-100 has black-gray-light gray, I'm assuming this to be a factor in achieving good neutrals in their profiles.

About all I can conclude is be ready to up the smoothness for profiles with questionable neutral performance.  All X-rite states in their help is that 50 should usually do the trick, higher values trade-off accuracy for "smoothness", however that manifests itself.  I read some earlier posts in LULA, not a whole lot of wisdom to be gained.  Anybody have a catalog of printers for which the smoothness adjustment is needed?

Richard Southworth

Added by edit - For what it's worth both printers are dye type.

[Doug Gray]

What settings are you using Richard?

If you could attach the generated CGATs file that has the settings embedded in the description. If you also attach the measured CGATs file I can look at it in some depth.

I don't have any dye printers but a single black is going to be difficult to get a good profile along the neutrals.

[rasworth]

Doug,

Attached are the spectral measurement file and my profile settings for the problematic profile.  The smoothness was changed to 80 to generate a satisfactory B&W profile.

Richard Southworth

[Doug Gray]

Doug,

Attached are the spectral measurement file and my profile settings for the problematic profile.  The smoothness was changed to 80 to generate a satisfactory B&W profile.

Richard Southworth

I've examined the file. It's a "tracked" measurement file. Looking at every 4th step in the staggered neutrals, which should, for a tracking file, be reasonably close to a*=b*=0 but it's not. Further, if I generate an BtoA (Lab to RGB) neutral curve the rgb substantially diverge. The smoother setting of  50 far more than the setting of 80.

This appears to be due to error offsets from the profile used to generate the tracking RGB values.
The main issue is that the file contains only 6 color patches. I understand the approach but highly recommend you put at least a "-m 4" or "-m 5" in the file used to create the tracking file. And at least "-m 6" in the final profile. There is such a large distance between the 6 color patches and the color info in the staggered patches, which  is noisy, doesn't convey enough info to make good tracking profiles even if the neutrals provided better info.

Also, I've noticed on my printers (all pigment) that, like your readings, there is quite a bit of anomalous noise and/or non-linearity in the patches at lower L*. I recommend using the -l 3 option to add finer grid points to mitigate this.

On my printer's I've also noticed a lot of variation between prints and am currently collecting data on these. The 9500 II is by far the worse. With L* varying depending on warm up or whether the printer goes through some sort of shaking prior to printing. This makes it difficult to determine the improvements of more patches as printer changes have to be factored out. I've got some techniques that does that when printing targets, basically by adding all the targets together then randomizing the patch locations before printing. The measured data is then descrambled and profiles are made from their respective sets.

However, I have to then run the profile verification test, which uses known colors, to make sure the L* changes that occur haven't been enough to alter the profile accuracy measurements too much. But, at least, what drift occurs on the verification affects all profiles being tested the same way.

[rasworth]

Doug,

As you noted it was a tracked file, but since I generated the first pass with the same target, other than the neutrals, I'm not surprised the neutrals did not track well.

Thanks for your input, I will in the future make sure there are more color patches than the six in this one, I was trying to be "efficient".  Apparently I created instability in the BtoA curve fitting, with the resulting divergence dependent upon the smoothness parameter.  I used this same target for my Pro 100 without seeing the problem, so I assume it's very much printer dependent.

I did find that for my neighbor's Pro 9000 Mkii I was able to extract better neutrals performance from a full color profile generated with 905 patches (i1Profiler auto generated) , by increasing the smoothness value.  Again, it seems that for this printer the smoothness increase is needed even with a large number of color patches.  Perhaps it correlates with the single black ink?

One thing I'm still not clear on is where to set the s parameter, any more wisdom to add for this item?  I've gone from 2 to 10 and in between, still don't have a good handle.  I understand I can tighten it for better behaved printers, but not sure how to assess behavior.

Richard Southworth

[Doug Gray]

Doug,

As you noted it was a tracked file, but since I generated the first pass with the same target, other than the neutrals, I'm not surprised the neutrals did not track well.

Thanks for your input, I will in the future make sure there are more color patches than the six in this one, I was trying to be "efficient".  Apparently I created instability in the BtoA curve fitting, with the resulting divergence dependent upon the smoothness parameter.  I used this same target for my Pro 100 without seeing the problem, so I assume it's very much printer dependent.

I did find that for my neighbor's Pro 9000 Mkii I was able to extract better neutrals performance from a full color profile generated with 905 patches (i1Profiler auto generated) , by increasing the smoothness value.  Again, it seems that for this printer the smoothness increase is needed even with a large number of color patches.  Perhaps it correlates with the single black ink?

One thing I'm still not clear on is where to set the s parameter, any more wisdom to add for this item?  I've gone from 2 to 10 and in between, still don't have a good handle.  I understand I can tighten it for better behaved printers, but not sure how to assess behavior.

Richard Southworth

I'm currently testing a panel of non-tracking profiles with "-m" set at 4, 5, and 6 and "-s" at 4, 8, and 12. This produces 9 separate profiles. To find the best performing set I duplicated all the patches, scrambled them, and printed out targets. Scrambling them all spreads out any printer difference between printed pages so will let me compare apples and apples. Duplication helps identify anomalous readings and reduces the noise 30%. They are currently drying. I plan to generate 9 profiles from these conditions and test which produces the closest neutral tracking. While the program defaults work well, I'd like to get the initial target as small as possible while still producing good tracking.

[rasworth]

Doug,

Looking forward to seeing your results and recommendations.  For me "magic" maximum patch numbers are 405 (i1Pro default on letter size) and 312 (i1Pro default on 8x10), which work out to -m at 6 and 5 respectively.  I'm assuming the -n setting is relatively non-critical for the non-tracking profile, as long as it is >= 60.

Question - do you foresee using the non-tracking profile results to determine a best choice -s for the tracked profile?

Richard Southworth

[rasworth]

Back to my rant about the i1Profiler smoothness setting.

I dug out a profile I made a year ago for my neighbor's Canon Pro9000mkii dye printer.  I extracted its data into i1Profiler and re-profiled, with no change to the profiler settings other than moving smoothness from 50 to 80.  The attachment image shows neutral curves for the two profiles, older on the left and newer (smoothness = 80) on the right.  Test prints confirm the newer profile prints with much better neutrals, i.e. no visible green tint.

The 910 target patch set was i1Profiler auto generated.  I have to believe there is magic, maybe a bug, with the smoothness setting for some printers, even with a significant number of color patches.

Richard Southworth

Correction - 905 patches, one of those patch counts that maximize the number of neutrals.

[rasworth]

I downloaded the Argyllcms package, in order to get some sort of comparison to i1Profiler.  Extracted the spectral data from the Pro6000 profile of the last post, converted to Argyll format with txt2ti3, and fed it to colprof to generate a profile.  If I left the -r parameter at default I obtained neutral curves similar to the i1Profiler version with smoothness at 50 (attached image).  Increasing -r in Argyll to 8 x default (4.0%) resulted in convergent neutral curves very close to that from i1Profiler with smoothness set to 80.

My understanding of increasing the -r parameter (from the Argyll help files) is it will smooth out the effect of higher patch read variations, either from a less consistent spectro or from a device with inherently more variability.  I tend to suspect the Pro6000, I've achieved good results using my i1Pro on another printer in the same time frame.  I realize it's a leap of faith to equate the iProfiler smoothness setting with the Argylcms -r parameter.  In any event I've soaked up about all the wisdom I can handle from this effort.

Richard Southworth

[Alan Goldhammer]

My understanding of increasing the -r parameter (from the Argyll help files) is it will smooth out the effect of higher patch read variations, either from a less consistent spectro or from a device with inherently more variability.  I tend to suspect the Pro6000, I've achieved good results using my i1Pro on another printer in the same time frame.  I realize it's a leap of faith to equate the iProfiler smoothness setting with the Argylcms -r parameter.  In any event I've soaked up about all the wisdom I can handle from this effort.

Richard Southworth

I use an i1Pro and have never seen a need to use the '-r' control.  Most of my errors are 0.3 and smaller and only see an occasional outlier which is usually the black patches near 100% black (I run a 51 step B/W patch set in the final profile reading which is probably overkill).  I tend to use standard tests prints for profile validation more than tools such as profile inspector.

[rasworth]

Alan,

I never used other than the default i1Profiler smoothness value, until I ran into this particular printer.  It's 10 years old, uses a single black ink, and no telling how much it has "moved" from its beginning point, although it still cranks out proper nozzle checks.

I've just started with Argyll, so I can't claim any real knowledge/experience wrt to the parameter choices.  I'm not sure it's as much an outlier issue, i.e. the printer produced a very different patch for the same input conditions, as it is "jumpiness" as it progresses thru the neutral region.  And I may be adding 2 + 2 and getting 5.

BTW the free portion of i1Profiler, target generation and measurement, is a nice front end to Argyll, based upon my very limited experience.

Richard Southworth

[Doug Gray]

Richard,

I'm in the process of completely revising the "tracking" methodology. Turns out the staggered RGB near neutrals sometimes are, in a mathematical sense, poorly conditioned. Depending on the printer's ink response, it can produce significant anomalies. This was also showing up in the work I am doing on synthetic profiles for ABW printing. But in looking at your CGATs files and profiles created from them, it really pops out and the approach needs significant modification to assure reasonable conditioning. This requires changing both the initial target used to create the tracking info and the second which incorporates well conditioned patch values.  I'm getting excellent results from the first change and expect the tracking near neutrals, which will be selected to provide good conditioning for the BtoA LUTs.

[rasworth]

Doug,

I look forward to exercising your next generation of tools.  I seem to have a worst case printer next door.

Richard Southworth

[Doug Gray]

Richard,

Thought I'd share some intermediate work since it should do quite well for your printer profiling.

I've attached a target CGATs file optimized for B&W. It's not a tracking target so it can be used with any printer. It has a very large number of near neutral patches (469) as well as a smaller number of color patches for a total of 685 patches. It makes an adequate color profile but one with really good B&W performance. It's as good as best tracking profiles but without their printer sensitivity. It should work very well for the Canon 5000. It produces average dE performance of .35 across the neutrals and .55 for color patches on my P1000.

[rasworth]

Doug,

Thanks, not sure when I'll make another run but I'll certainly report back when I do.

Richard Southworth

[rasworth]

Doug,

Curiosity got the better of me, had to plot out your latest patch set.  I'll be interested in hearing your rationale for the interlinked 3x3x3 groups along the neutral axis.

Richard Southworth

[Doug Gray]

There are a few reasons for this.

1. dE 2000, unlike dE 1976, better describes visual sensitivity to small changes in hue and saturation near the neutral axis. And it better describes the very reduced sensitivity to increases in saturation with more intense colors.

2. Printers vary in how far, and how rapidly, they diverge from neutral when printing the range of device neutral RGB values. To get the highest accuracy, one needs to have a reasonable number of nearby, near neutral patches in all directions.

3. For the Canon dye printer you provided a CGATs for, and the Pro1000 I have, there is a rapid change in a* and b* of 5 to 8 units over the range of L*=3 to 12.  The LUTs I1Profiler makes have steps of about 3 for L* but steps of 7 for a* and b*.  But at low L*'s dE 2000 is 2 to  times more sensitive to a 1 unit change in a* v a 1 unit change in  L*.  To get good neutral tracking at these darker areas, a high density of patches is needed. I expanded the area around the neutral axis so that these variations can be better measured in this region.

4. For the Pro1000, this high density is needed only for L* < 20 but for my Epson 9800, a higher density is needed throughout the range. While the 9800 deviates from the neutral axis much less than the Pro1000, its deviation is more lumpy throughout the range. It's worst case is in the neighborhood of  80 < L* < 90.

So this set of RGB values generates good neutral performance with both but at the cost of a lot of near neutrals.

Currently, it appears some sort of variable density/step distance approach would work best for making a tracking profile generated from both the deviation from neutral as well as the lumpiness on the first pass profile without needed 400+ near neutrals. But in the meantime, the high, near neutral set works very well.

[rasworth]

Got it, thank you.

Richard Southworth

[rasworth]

Doug,

To get a better handle on your new patch set I "dumped" the rgb values into one of my better profiles for the stubborn Pro9000 printer, graph attached.  Even though the neutral curves for this printer droop significantly there was still enough spread on your near-neutrals to completely enclose the L* axis.  I therefore assume this patch set would be very good for creating training profiles?  Maybe overkill?

Richard Southworth

[Doug Gray]

Doug,

To get a better handle on your new patch set I "dumped" the rgb values into one of my better profiles for the stubborn Pro9000 printer, graph attached.  Even though the neutral curves for this printer droop significantly there was still enough spread on your near-neutrals to completely enclose the L* axis.  I therefore assume this patch set would be very good for creating training profiles?  Maybe overkill?

Richard Southworth

Richard, the patch set should produce very good, final, profiles for B&W, but average to marginal color performance. It's not intended for creating tracking profiles per se, but rather giving me more detail in designing a better tracking, two step process. I may include an option to just add it to the good color packed grids with large -m and other settings.

I posted it for you to use because it should make excellent B&W profiles for your colleague's printer.