Z3200 - Better ICC Color Profiles

[Brad P]

Very interesting. I’ve been playing around with B.B.’s all morning, including rewatching Andrew’s excellent video, playing around with the saturation slider, looking at results on my Adobe RGB monitor and comparing the monitor to actual prints.  I think I finally get how that all works together now.  Thanks. 

One thing I especially believe I’m learning is that I might get more at this point looking for optimal profiling software for my setup and lighting conditions rather than increasing patches from 1728 to 6000, although that would help too as Mark and Mark are discussing in a new post on using the Z to print and measure a 6000 patch set.   And as helpful as soft proofing is in the workflow, it’s absolutely necessary to look at actual prints in reaching decisions there. 

[Ernst Dinkla]

Hi Ernst - thanks for that - it helps fill in puzzle pieces. Do you recall what was the last year they upgraded APS?

You also mentioned that the Software would still make an ICC profile without the dongle.  That's interesting.
Would like to get my hands on a copy and the upgrades.

So MHMG and I are making some good progress - will keep everyone posted with the results eventually.

Thanks Ernst -

Mark

This morning I tried to reinstall the HP-APS software based on older installation files I kept. The registration keys did not work anymore and neither the activation page at Gretag Macbeth now X-rite or the registration page at HP exist anymore. The Display Two puck possibly went kaput or I have the keys from the other puck here, the one I sold with the other APS license. I doubt the last though as I had APS working after the sale. Never used the puck anymore for the monitor so it was really a dongle. The Display Pro is much better.

Making profiles based on measurement files looked possible but at the last stage the same, my words: 'wrong Display etc activated, use the other one' error appears. So no route there. I have to dig in older archives but it is probably easier to take the ArgyllCMS solution now. I contributed for that in the past and bought the Android Argyll app last year so I might as well use that.

With the last version 141 of the HP-APS  (7-2011 is the likely date of launch of that one) I see 'Copyright Logo GMBH Steinfurt' in the Z3100 profiles made and 'Copyright X-Rite' in the Z3200 profiles made. Structure of the data seems to differ too. Color Center profiles have the HP copyright in the headings.


Met vriendelijke groet, Ernst

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

[Mark Lindquist]

Thanks Ernst, more puzzle pieces with which to solve the missing APS mystery.  MHMG and I both think it became a licensing issue, once everything moved to i1, and your recent information helps to confirm that supposition.  Man, I’m sorry you can’t get your APS to activate.  I just found Michael Reichmann’s original article on making profiles with the Z3100 and APS:

Michael’s Article on APS Profile Making on Z3100

(This is part of the subscription based archives) So there are many interesting tips and tricks for using the APS and the means to trick it into making larger targets.  It’s interesting, especially considering it goes back to 2009.

But I agree, Ernst, the current color center does most of what APS did, and no tricks, with the exception of making the final ICCProfile from the .TXT file.  Argyll works really fine, especially if you are using Rel Col. I imagine Terminal is a second language for you, but if you are interested, you can use the free  GUI my son built:

ICC GEN  (WIN / MAC)

FWIW, Argyll and Drop RGB perform almost identically with RelCol.

Not sure if you followed this, but MHMG and I both experienced a bug in the Mac Utility where it refuse to finish printing targets over 2000 patches, yet goes on to measure then fails.  I found that Windows will, in fact, sail right on through, right up to 6000 patch targets, flawlessly.

I do think the APS is now history, but for those still using it, it does provide a model for doing the same thing in color center.  Once in the color measurement dialog, when choosing the target size, instead, just click the + sign which takes you to an opportunity to add a chart file to the list.  It adds it, but does not show it until you go back to the drop down menu, and sure enough, it is there.

Chart files made with i1Profiler2, modified via Doug Gray’s suggestions work great.

Maybe it’s just a matter of finding the right key, but if you can’t get your APS working I don’t think it’s a big loss at this point.

Best,

Mark L

[Ernst Dinkla]

Thanks Ernst, more puzzle pieces with which to solve the missing APS mystery.  MHMG and I both think it became a licensing issue, once everything moved to i1, and your recent information helps to confirm that supposition.  Man, I’m sorry you can’t get your APS to activate.  I just found Michael Reichmann’s original article on making profiles with the Z3100 and APS:

Michael’s Article on APS Profile Making on Z3100

(This is part of the subscription based archives) So there are many interesting tips and tricks for using the APS and the means to trick it into making larger targets.  It’s interesting, especially considering it goes back to 2009.

But I agree, Ernst, the current color center does most of what APS did, and no tricks, with the exception of making the final ICCProfile from the .TXT file.  Argyll works really fine, especially if you are using Rel Col. I imagine Terminal is a second language for you, but if you are interested, you can use the free  GUI my son built:

ICC GEN  (WIN / MAC)

FWIW, Argyll and Drop RGB perform almost identically with RelCol.

Not sure if you followed this, but MHMG and I both experienced a bug in the Mac Utility where it refuse to finish printing targets over 2000 patches, yet goes on to measure then fails.  I found that Windows will, in fact, sail right on through, right up to 6000 patch targets, flawlessly.

I do think the APS is now history, but for those still using it, it does provide a model for doing the same thing in color center.  Once in the color measurement dialog, when choosing the target size, instead, just click the + sign which takes you to an opportunity to add a chart file to the list.  It adds it, but does not show it until you go back to the drop down menu, and sure enough, it is there.

Chart files made with i1Profiler2, modified via Doug Gray’s suggestions work great.

Maybe it’s just a matter of finding the right key, but if you can’t get your APS working I don’t think it’s a big loss at this point.

Best,

Mark L

I will manage, not really a thing to worry about.  Some years back (2011 and later in the files data) I tried to add grey patch targets to Color Center but later on went for APS and had a trial 17 patch greyscale target to print and get measurements for the QTR B&W profiler.  Z3200 in B&W mode to use it on. And also on a desktop printer with a trial quad ink mix (Z3200 MK + PK + dilutions). With a new manual spectrometer it was easier to do it manually, 34 patches target file I ended with. I think for the Z3200 it will be better to stay in color mode for print and color management as profiles with enough grey patches will solve the a b deviations easier than ink mixing does. The MK ink is warmer than the neutral PK and Grey inks. Count to that OBA and non-OBA paper whites and it becomes hard to get that a b controlled in ink mixing. So I am grateful for all the puzzling done by the people that contributed to this thread.

Not much of a programmer, more the trial and error guy in manipulating CGAT files. So I recognized the patch number change needed and what columns to strip for a profile creator acceptance.


Met vriendelijke groet, Ernst

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

[Brad P]

Well, I had some extra HP Universal gloss paper lying around . . .

I found a work around to print a 4357 target without leaving Mac.  Who knows, it could go up to 6001 or infinity.
   
It's 9:20 pm local time and I'm done for tonight having just printed it.  Will let it dry and scan in tomorrow morning.  I'm guessing the scanned file might drop into DropRGB or be able to be processed in Argyll, probably also i1Profiler since it's based on a color chart there and only has the HP header.  Here's the procedure.

1.   Open i1Profiler
    a.   In User Mode, click “Advanced” 
    b.   Under Printer, device selection, select “RGB Printer”
    c.   Under Workflow, select Profiling
    d.   Under Smart Patch Generator, select 4357 (or another profile target patch number)
    e.   Click on Save
    f.   In the save dialogue box,
           i.   Name the file (I named it Chart 4357 Patches}
           ii.   Select the directory where the file will reside.  Desktop works
           iii.   Select “Custom CGATS(*.txt)”
           iv.   Then click on “save”
    g.   A pop up appears with a lot of custom CGATS options (including a D50 illuminate default selection).  Accept the defaults.  I’m think I’m generally right to do so, but someone might want to double check me on that.

2.   Open the file in Textedit.  Swap out the header information for this one and save.  I modified this header after an APS profiling chart embedded deep in in the APS application files (I too lost my dongle, and sadly remember a tiny bit of DOS from 1989ish.  The header will only work with a file of equal size as this outlined one.  I suspect it has something to do with the first line).  All spaces are actually tabs.

LGOROWLENGTH   21
CREATED   "12/10/2017”   #Time:   12:57
KEYWORD   "SampleID"
KEYWORD   "SAMPLE_NAME"
NUMBER_OF_FIELDS   5
BEGIN_DATA_FORMAT
SampleID   SAMPLE_NAME   RGB_R   RGB_G   RGB_B
END_DATA_FORMAT
NUMBER_OF_SETS   4357
BEGIN_DATA

3. Open  the HP Utility to print the test chart
    a.   Select the printer, Paper Preset Management, the paper to be profiled, the little gear box on the bottom, then color measurement.
    b.   Select Print a color chart for later measurement,  “Print a color chart for later measurement” and (if you use it), check Print using gloss enhancer, then select Continue
    c.   In the “Select Color Chart” menu, click on the plus sign, select the file just created with i1Profiler as modified, click on Open, go back to the drop down menu and select that file.  Make sure it’s selected in the “Select Color Chart” dialogue box then click Continue. 

The chart should begin printing.

Badly dim lit iPhone 6 pics attached off a 36" roll.

ADDENDUM:  Proof of concept.  I waited an hour, scanned it and saved a CSV file.  Dropped that into a BasICColor DropRGB demo and it put out an ICC file.  Attached the resulting ICC file to the Universal profile in the utility, and voila.  It works.

Looking at the screen in P in soft proof mode, the BB tests look better.  But it's a different paper (a significantly worse grade, HP Universal Glossy 36 inch instead of Ilford Gold Fibre Gloss 44).  The print looks very good for the Universal paper except for the newly added Granger rainbow which is a mess.  The BW ramps look clearly better in person in the new print.  For now, I'm blaming it all on that paper and in the picture warm old dim LED living room lighting.  Dunno. I'll do more tests later but for now, I'm happy.  ColorThink tests on the results with Argyll and other engines will have to wait (12:40 am), probably for a few days for a new roll of IGFG to arrive.

Photos attached.

ADDENDUM 2, strangely this did not work when repeated on a different paper profile.  In fact, the target file would not even load into the HP Utility.  I’m at a loss why the different results. 

[Mark Lindquist]

Hi Brad,  just to confirm, yes this is the exact work flow MHMG and I have been discussing and Mark MHMG has been working on.  So yes, you have the correct work around for Mac Utility.

As far as adding the patch target size file in under the list, clicking the “+” sign is the correct method, and then using the drop-down menu, the new target will show up.  Install profile is the way to go, as Geraldo pointed out in his original thread in order to get the newly generated /ICC file included in the custom paper list.  Thanks for listing the steps.  We’ve been doing this, but haven’t posted it yet, wanting to finalize our test results.  I ran into a glitch with the Windows HP Utility, where although it would print and measure the 44” wide 6000 patch target, it failed to print and measure the 4357 target on 24” Wide Moab Entrada.
After much screwing around, I put a 6 ft. Long table in behind the printer, cut the roll into a sheet that was about 5+ feet, and using “load paper with skew check”, was able to get the spectro to read the chart using the very same workflow of “measure a previously printed chart”.  MHMG is testing the bugginess of printing tiff files with Apple Color Sync, etc., and we are continuing to carefully plod through various iterations until we know for sure what bugs are and aren’t in both the Mac and Windows HP Utilities before publishing the results.  Your list of steps will come in handy when we do come to that point.  Very good. It also confirms our results.

Mark

[MHMG]

... I found a work around to print a 4357 target without leaving Mac.  Who knows, it could go up to 6001 or infinity. ...
   

Hi Brad, glad you succeeded printing and measuring the 4357 i1P generated target directly from the HP color utility on the Mac. Our workflow is AFAIK identical except for your modification to the header of the patch count target. That said, I have my doubts this is the source of my Mac failing to print the whole patch set direct from the HP color utility for two reasons:

1) I've had the 1728 patch target supplied in the HP Color Utility by HP also fail in the same way at least once although it succeeds usually. Surely that patch target has the header info exactly as it should be (I'd check that out, but I'm not sure where the HP utility software hides that file).

2). I can take the 4357 and other  i1P exported CGATs chart files  directly into the HP Color Utility, and the HP color Utility knows exactly how to lay them out with no truncation when it export a tiff file one can print and measure as an Image on the HP or other printers. It seems to me that if the HP is getting confused by the header info in the custom target, it would fail to parse the chart file correctly when laying out the patches for a tiff file export. Yet the exported tiff image files are generated exactly as they should be.

That said, I will take another stab at printing the 4357 target using your suggested header modification hopefully later today.

best,
Mark McCG
http://www.aardenburg-imaging.com

[Doug Gray]

Cautions on using CGATS files with I1Profiler.

Major detailed breakdown of I1Profiler handling of fractional RGB values:

1. It does retain them if generated and the "smart patch" generator does generate them. However, stored data I the profiles truncates fractional RGB info.
2. It does round RGB values when printing or creating target TIF files. However, the fractional values are used in calculating the profile.
3. Sampled data can be saved in the profile. Unfortunately, fractional RGB values are truncated in the saved data. If that data is re-used, expect the profiles to have additional error due to the truncation.

For consistency in all modes, I would highly recommend rounding all target CGAT files that have fractional RGB values before using them in any target making/reading application.  This retains optimal compatibility with I1PRofiler.


I1Profiler may, under certain circumstances, use fractional RGB data when calculating profiles. If you load a CGATS file with fractional RGB values and spectral data it will calculate a profile using that fractional data.

However, I1PRofiler discards fractional data when you just load targets. Further, it truncates rather than rounds so a value of 89.9 becomes 89. Not using fractional data is necessary with I1Profiler given that it prints out to the device driver with 8 bit values.  Also, if you save a tiff file it's in 8 bits too.

This can introduce small errors when using targets with fractional RGB data if the printer utility doesn't have a way to deal with them and just prints 8 bits. Especially if the scanned CGATS file contains the fractional data.

[Brad P]

Thanks Doug.  Just in time, I found some time and was just about to try to print some up.  Now I'll hold off a bit

Looking at all the i1P generated CGATS files, all the RGB data are solid single digits (e.g., 204.00) instead of some files I saw at least in CMYK targets that contain decimals out to the fourth digit.  Is this single digit rounding error the source of your concern?  I could understand that a small patch set might not be so much affected by that, but the kind of detail we're trying to render here would.

If that's correct (or regardless), is there some other CGATS rendering software or website you could refer us to that does a better job?  I was actually thinking of making a spreadsheet set of data and plugging that in, but I'd rather find something more accessible.

[Doug Gray]

Thanks Doug.  Just in time, I found some time and was just about to try to print some up.  Now I'll hold off a bit

Looking at all the i1P generated CGATS files, all the RGB data are solid single digits (e.g., 204.00) instead of some files I saw at least in CMYK targets that contain decimals out to the fourth digit.  Is this single digit rounding error the source of your concern?  I could understand that a small patch set might not be so much affected by that, but the kind of detail we're trying to render here would.

If that's correct (or regardless), is there some other CGATS rendering software or website you could refer us to that does a better job?  I was actually thinking of making a spreadsheet set of data and plugging that in, but I'd rather find something more accessible.

Here's entry 17 of the smallest smart patch size (400) generatable in I1Profiler:
85.00      72.86       0.00

This is printed as an 8 bit tif containing:
85.00      73.00       0.00

But this is what's saved in the generated profile
85.00      72.00       0.00

So, sadly, there is no simple way to round the RGB data with i1Profiler. I have various workarounds but hadn't previously noticed these details. Just knew I1Proffiler messed things up in places if there was fractional RGB data.

I think I've make a few profiles from extracting the saved data in a profile and it turns out that is a big mistake.  Never use saved data in a profile UNLESS it had no fractional RGB in it at the start.  That alone is a compelling reason to use only RGB data w/o a non-zero fractional component.

[digitaldog]

So, sadly, there is no simple way to round the RGB data with i1Profiler.

Does it matter? Can you SEE the 'errors' on output?

[Brad P]

Does it matter? Can you SEE the 'errors' on output?

Both of you are much better than me to opine on that, but I’d imagine since it’s +/-0.86 of the green channel, the right answer is either no or barely, at least for that one patch.  This single observation causes me to wonder, Doug, how pervasive these rounding errors are and the fallout from that the cumulative, limiting effect of these rounding errors on the efficacy of larger patch target sets. Very possibly not much at all, but it seems better to manage the things we can when we see them.  Which leads me back to an earlier question, is there a better tool to generate the RGB data than i1P?  It looks like some of the $800+ profiling tools might generate some, but at unknown quality (to me) and at that cost, the i1P solution looks probably good enough. 

[Doug Gray]

Both of you are much better than me to opine on that, but I’d imagine since it’s 0.86 of the green channel, the right answer is either no or barely, at least for that one patch.  This single observation causes me to wonder, Doug, how pervasive these rounding errors are and the fallout from that the cumulative, limiting effect of these rounding errors on the efficacy of larger patch target sets. Very possibly not much at all, but it seems better to manage the things we can when we see them.  Which leads me back to an earlier question, is there a better tool to generate the RGB data than i1P?  It looks like some of the $800+ profiling tools might generate some, but at unknown quality (to me) and at that cost, the i1P solution looks probably good enough.

That was just an example of a color showing how it was rounded when printed and truncated when saved in the icc profile. Most of the patches have fractional values. But is rounding them going to produce visual differences? Almost always they will not.

Still, it's something I've avoided for a long time. When I make a patch set that has fractional RGB values I've almost always rounded the RGB values as the first step before using it as a target. This largely as a result of observations by either Ethan or Mark M. I just wanted to avoid any issue.

However, now that I knowI1Profiler actually uses fractional values I may alter my process. I have the ability to create 16 bit tiffs with those fractional values that are compatible with Isis.

However, Andrew's point is germane. At best this is a small affect and very unlikely to create visible differences. I've investigated the impact it has on measured profile accuracy and it's a small effect until one is making profiles over 2 or  thousand patches. I suspect it may be more important with very high ( Over 5k) patch count profiles because these errors, which are small but fixed, become a larger portion of the distance between adjacent patches.

It's more of academic interest than anything else. While I can measure improvements with patch counts over 2k I can't really see any difference in the prints. I can't visually tell any difference in reference prints. Even with the standard 957 count target it's difficult to see any difference. Possible, but just barely.

Being a nitpicker, I just don't like these little discrepancies.

[Brad P]

Thanks Doug.  Very helpful and, for us nitpickers, interesting to know. 

[Mark Lindquist]

It's more of academic interest than anything else. While I can measure improvements with patch counts over 2k I can't really see any difference in the prints. I can't visually tell any difference in reference prints. Even with the standard 957 count target it's difficult to see any difference. Possible, but just barely.

Being a nitpicker, I just don't like these little discrepancies.

I found that in smaller prints it was difficult to see differences in identical 17” prints printed with 1728 vs. 4357 or 6,000, but much easier to tell the diferences in 44” prints.  The larger the print the more apparent the differences by eye.  Subtle differences, but there, nevertheless.  At that point it seems to become somehow that the realm of “wine connoisseur” descriptors must enter in to be able to discuss those differences, however. Words like smoother, buttery smooth, snappier, more open, richer, more full, etc., become part of the vocabulary, as there is little or no other method of describing differences which are there, but not always immediately apparent, especially without having the prints side by side to compare.  I reasoned that if the print at 44” looked really great, (actually spectacular), AND, a 6000 patch profile target was used to print it, why then would there NOT be a difference, meaning “if it looks, smells, walks, talks and tastes like a duck,” as the old saying goes, “then maybe it is a duck.” 

Once one becomes attuned to looking for differences, and begins to become aware of what to look for in the subtle nuances of print variations done under controlled printing methods (printing the very same file but with 2 different profiles, easily accomplished with the Z (under job queue > change loaded paper, reprint), then I believe we are looking at an improvement.  Talk to audiophilles, and you will hear the same kind of rhetoric - the waxing lyrical, attempting to discuss sound quality nuances of more expensive equipment. 

Thus we have entered into the realm of the law of diminishing returns.  It may be easy to achieve 85% of any given endeavor on a scale of “perfection”, yet with every percent greater acheivement, the proportionally greater requirements of resource required to attain improvements, until finally, the law of diminishing returns has its way until eventually a paradigm shift renders the point moot. I know this to be true in metal machining and in robotics.  Getting 85% to the goal is likely doable, but beyond that, it becomes like pulling teeth to make significant advances.

Perception of quality is of a subjective nature and becomes about taste and ushers in the potential for snobbery.  But, apparently, it is the way the world works. Connoisseurship is very real, and among so called connoisseurs, these subtle nuances define the variations in “quality”.  I agree, to some extent, being one who chases the rabbit down the rabbit hole and through the warrens, in many differentt areas.
Yet this I have come to understand:  Perfection is the enemy of excellence.  Obsession the teaser of sanity.  To each his/her own.  Do what floats your boat, follow your bliss.  Or be practical, or frugal, or sane, whatever is the opposite.  Make prints your own ways, according to your own choices.  It’s not that making a 6000 patch target ICC profile doesn’t come at considerable cost....

And... understanding and communicating those differences as well....

Best,

Mark

[MHMG]

Hello everyone,

I'm returning to this thread after a couple of days where I had to attend to other things. Here's my latest thinking on using the Z3200 spectro with big targets and to a more general quest of building more accurate profiles.

1). I did try Brad's suggestion of modifying the header in the 4357 patch file, and although Brad successfully printed and measured on his Z, I still had the same bug (symptom being the HP utility thought it printed successfully, but in fact the printed patch count fell short). This remains and unresolved issue for me and may well be related to the fact that I'm on High Sierra with latest HP color utility version (who knows?).

2). Doug's discussion of rounding issues with the i1Profiler data sets is duly noted. Here's my take on it. If you go into PS and use the info tool (it rounds to whole integers when showing both RGB values and LAB values, one can try moving different triplets up or down just one unit of RGB in one channel, say for example raising G from 50 to 51 while holding R and B at 50. Try enough examples and we can see a* and b* values going up or down by 1 unit in many instances, hence 1dE or less depending on the chosen RGB triplet. For moderate chroma to vivid color chroma values, this is indeed probably nothing to worry about and will be mitigated by the profiling engine algorithms, but when it comes to neutrals and near neutrals, say two gray patches sitting side by side in the viewer's field of vision where one patch drifted one up in a* or b* while another drifted down in a* or b*, the errors could become just noticeable to the discerning viewer. Hence, it's worth looking at the rounding issue a little more closely as we try to create better, ie., smoother and more accurate gray ramps on full color printers.

3). This may seem counterintuitive, but Xrite's decision to truncate (i.e. round down) rather than round to nearest whole value has some merit, or at the very least is probably no worse than round to nearest whole value.  Consider a chart text file that has one patch with one RGB channel equal to 15.51 and another at 16.49. Round down will cause the printer to print two patches, one with 15 as output value and the other with 16 as the "no color adjust" input values. Round to nearest whole integer will create and 8bit tiff untagged image with 16 and 16 respectively for those two patches. As long as the profiling engine is respecting those RGB text file value differences when using the corresponding spectro measurements to build the profile, I think the outcome is similar, perhaps even better with round down because round down will attempt to preserve differences in the RGB triplet values consistently whereas round to nearest value will cause some RGB triplet pairs to take on bigger differences while others take on smaller differences.

4). All that said,  given that fact that i1Profiler generates 8bit tif target images to be printed and measured and HP color Utility does the same, it is certainly easy to avoid the uncertainty of the rounding issues through the printer output and target measuring pipeline, simply by following Doug's suggestion to clean up the patch chart test file, so that the printer is getting fixed integers with no rounding errors when printing the target in a "no color adjust" printer pipeline, and the profiling engine is also getting those corresponding integers when trying to "smooth" and build the resulting ICC profile.

I'm going to take a breather for now from trying to solve the printer bug issue in the HP Utility, and will now try a "straight print" approach by exporting an 8bit patch target tiff image and letting Apple ColorSync Utility "print as target" with 100% magnification. It's imperative for the Z3200's measurement step that the patches are printed at 100% magnification, otherwise the measuring step is bound to fail. There's also a trick to getting CSU to print 100% precisely, but I will cover that step later when Mark L. and I eventually have a chance to put together a tutorial on how to print these big patch count targets reliably on the Z3200, and what the ramifications are for better profile quality (if any) as patch counts get higher and also include more gray/near gray patches than the venerable 1728 (12x12x12 grid) classic target.

cheers,
Mark
http://www.aardenburg-imaging.com

[mscherlacher]

Hello all,

I experienced a strange little bug when saving the 4357 patch file out of i1Profiler. When I opened the file to inspect it the Number Of Sets said 4357 but the actual saved sets only went to 2000 something or other (I deleted it without noting the actual number). I saved it again and it was correct this time around.  I have not been able to repeat this and perhaps it was something I did wrong but I have no idea what that could have been. Thought I'd throw this out there, just in case this is what is happening to you too.

Thanks one more time for sharing all of this, I'm learning a lot and making better prints because of it.

Mike

[MHMG]

Hello all,

I experienced a strange little bug when saving the 4357 patch file out of i1Profiler. When I opened the file to inspect it the Number Of Sets said 4357 but the actual saved sets only went to 2000 something or other (I deleted it without noting the actual number)....

Mike

Mike, that would have been a good and "why didn't I think of that!" reason for the printing bug I've been experiencing, but I checked my 4357 chart files, including the one that had Brad's suggested header modification, and all have complete data sets.

Onward and thanks,
Mark
http://www.aardenburg-imaging.com

[Ethan Hansen]

Regarding i1Profiler data truncation: The last time I thoroughly vetted the consequences of using fractional data in i1Profiler was with version 1.63 (late 2015 vintage). I dug through my Excel archives and here were the conclusions.

I ran several cases:

• The default i1P 8-bit target image with RGB values rounded to the nearest integer paired with the 16-bit reference. This is the default i1Profiler configuration.
• Default 8-bit i1P target image with reference file truncated to 8-bits (i.e. values truncated down to the nearest integer).
• Default 8-bit i1P target image with reference file rounded to the nearest integer.
• 16 bit target image, 8-bit reference file rounded to the nearest integer value.
• 16 bit target image, reference file truncated to 8-bits.
• A true 16 bit target image having the fractional values in the measurement reference file (target image 16-bit, reference file 16-bit).
• A version with both image and reference truncated to 8 bits.

Options #1, #2, and #3 all made similar but not quite identical profiles, although the differences barely exceeded the measurement tolerances of an iSis. The target images were the same for all three tests. If i1Profiler truncates or rounds 16-bit reference values prior to the profile calculations the profiles should differ. The differences were statistically significant at only a 90% confidence level; these results don't definitively prove i1Profiler makes use of 16-bit reference data. A simple statement by X-Rite would be greatly appreciated.

Options #4 and #5 produced profiles that were statistically different at a 95% confidence level both from each other and from the previous three tests. Having different target image values ended up producing a larger effect than fractional differences in reference files.

Finally, options 6 and 7 - where the target image and reference files should be matched exactly - gave profiles with outputs with no statistically significant differences. This again hints that i1P makes use of 16-bit reference values.

We went on to quantify the profile accuracy by printing a target through each profile and measuring the resulting colors vs. predicted values. The winner was the final option above, with matched 8-bit target image and reference file. Average dE2000 was 0.68 on a Fuji Frontier contone printer and 0.91 on an Epson inkjet. The other options gave Fuji errors in the 0.95 - 1.32 range and Epson errors of 1.18 - 1.67 dE2000.

The larger question is whether these mathematical gyrations made and perceptible difference in the final print. For the most part the answer was no on both printers. The exception was in B&W images where the clear winner again, particularly on the inkjet, was the profile made with matched 8-bit target and reference. Highlight transitions had visibly fewer crossovers and smoother transitions. The matched 16-bit image/reference profiles made prints that some of us thought looked better than from options 1-5 while the rest couldn't see any difference.

[MHMG]

I ran several cases:

• The default i1P 8-bit target image with RGB values rounded to the nearest integer paired with the 16-bit reference. This is the default i1Profiler configuration.
• Default 8-bit i1P target image with reference file truncated to 8-bits (i.e. values truncated down to the nearest integer).
• Default 8-bit i1P target image with reference file rounded to the nearest integer.
• 16 bit target image, 8-bit reference file rounded to the nearest integer value.
• 16 bit target image, reference file truncated to 8-bits.
• A true 16 bit target image having the fractional values in the measurement reference file (target image 16-bit, reference file 16-bit).
• A version with both image and reference truncated to 8 bits.

Ethan, it's good to know that item 7 was the winner because I don't see any way using i1Profller to do what you did in item 6, i.e., save out the target image as 16-bit. It seems to want to generate 8-bit only. Am I missing a preference setting somewhere in i1P, or did you use another piece of software to generate a 16-bit target image from the 16-bit reference file?

cheers,
mark
http://www.aardenburg-imaging.com