Pages: [1]   Go Down

Author Topic: Optimizing printing profiles from i1Studio Spectrophotometer  (Read 2899 times)

NAwlins_Contrarian

  • Full Member
  • ***
  • Offline Offline
  • Posts: 225
Optimizing printing profiles from i1Studio Spectrophotometer
« on: October 28, 2018, 02:36:08 pm »

I'm in the process of making my own ICC profiles for printing with a rented i1Studio Spectrophotometer (prior post: https://forum.luminous-landscape.com/index.php?topic=126835.0). The X-Rite software has a function to optimize profiles using supplemental patch sets generated from your own images--you load a TIFF or JPEG, and it generates an additional 20 or 30 patches or whatever.*

My questions are: (1) does using one of my own images to generate more patches and optimize the profile in any way make the profile less suitable for printing other images? I can imagine that if you 'push' the profile in one direction to make it better for one image, for other images the 'as pushed' profile may be less good than was the non-optimized profile. On the other hand, maybe adding any additional patches to what has already been printed, read, and incorporated can only make the as-optimized profile better for all images. And if that is the case, then (2) does performing successive optimizations using different images, at least up to a point, continue to improve the profile? I can imagine that it does, but I can also imagine that the software can only adjust through one set of optimization patches, in which case a second optimization would necessarily undo the results of the first optimization before incorporating the readings of the second optimizing patch set.

Any guidance you can give me on these points is appreciated. Thanks!

*The i1 Studio's basic function, or at least its software's basic function, is that it prints one standard target of 50 patches, you read those with the spectrophotometer, from that reading it generates a second, somewhat-customized set of either 50 patches (for color) or 100 patches (for B&W), it prints those, you read the second set with the spectrophotometer, and it generates an ICC profile.
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 20630
  • Andrew Rodney
    • http://www.digitaldog.net/
Re: Optimizing printing profiles from i1Studio Spectrophotometer
« Reply #1 on: October 28, 2018, 02:55:03 pm »


https://forum.luminous-landscape.com/index.php?topic=124207.msg1040447#msg1040447https://forum.luminous-landscape.com/index.php?topic=126392.msg1062187#msg1062187
1. I've used my images for part of the targets and yes, useful depending on what you pick (the auto selection from an image doesn't work very well).
2. Haven’t seen this and there's a point of diminishing return but of course, this depends on the number of patches and how they were created for the first optimization.
Logged
http://www.digitaldog.net/
Author "Color Management for Photographers".

NAwlins_Contrarian

  • Full Member
  • ***
  • Offline Offline
  • Posts: 225
Re: Optimizing printing profiles from i1Studio Spectrophotometer
« Reply #2 on: October 30, 2018, 09:41:39 am »

Thanks Andrew. Since I was short on expertise and using a rented device due to be returned, I just tried to optimize one profile, out of the twelve I made, in a relatively automated way. For whatever it's worth, although the X-Rite software for the i1Studio has separate processes for making color and B&W printer profiles,* it only lets you optimize color profiles. So I took the profile I'd made for Mitsubishi Pictorico White Film on the Canon Pro-100 and optimized it, letting it use the maximum number (30) of additional patches it chose for itself from one of my images. After printing, drying, and measuring those additional patches, it spit out the optimized profile. For whatever little it might tell you, across two different printers, all the B&W profiles are 1584KB, all the color profiles are 1602KB, and the optimized color profile is 1607KB. Unfortunately I have not had time to do any printing with these profiles, so maybe the most important test remains to be performed.

*Interestingly to me, the B&W profiles use more target patches, with the second, semi-customized target containing an additional 100 patches, instead of an additional 50 patches for the color profile second target.
Logged

nirpat89

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 661
    • Photography by Niranjan Patel
Re: Optimizing printing profiles from i1Studio Spectrophotometer
« Reply #3 on: October 30, 2018, 12:46:11 pm »

Thanks Andrew. Since I was short on expertise and using a rented device due to be returned, I just tried to optimize one profile, out of the twelve I made, in a relatively automated way. For whatever it's worth, although the X-Rite software for the i1Studio has separate processes for making color and B&W printer profiles,* it only lets you optimize color profiles. So I took the profile I'd made for Mitsubishi Pictorico White Film on the Canon Pro-100 and optimized it, letting it use the maximum number (30) of additional patches it chose for itself from one of my images. After printing, drying, and measuring those additional patches, it spit out the optimized profile. For whatever little it might tell you, across two different printers, all the B&W profiles are 1584KB, all the color profiles are 1602KB, and the optimized color profile is 1607KB. Unfortunately I have not had time to do any printing with these profiles, so maybe the most important test remains to be performed.

*Interestingly to me, the B&W profiles use more target patches, with the second, semi-customized target containing an additional 100 patches, instead of an additional 50 patches for the color profile second target.

I had made the standard profile for Canson Photo Rag on my P400, which did a pretty good job on my color images.  When I tried to use it on one of my monochrome images, which I tone to sepia-like color, the result was a noticeable green tinge in the shadows, almost like a split-tone between brown and green, not what I was going for.  The print also seemed a little bit darker.  So I tried the optimization process by feeding a 101-steps step-wedge toned identically as the monochrome images.  The resulting profile gave a print that was free of the earlier greenish cast and the shadows seem to be separated better as well, slightly lighter overall that matched the display image much better than before.  So it looked like the effort paid off in this one case for me.  Just to check, I printed one of my other regular color images and it looked like it had tiny bit of more magenta as well a little lighter as one might expect - not a dramatic difference but not sure if that was an improvement over the first profile either. 

My thinking is that the optimization is only (or most) valid for the image that you feed in or the images that have similar color gamut to that image.  But it does not make the profile "better" overall for all images.  If successive optimizations are done on top of each other with different images, I would guess it would be best for the last image, not asymptotically better towards the ideal for the whole gamut of the printer.  Granted my experience is a sigle case of the extreme, so it has limited value for any generalizations.  Perhaps someone at X-rite can throw some light on the subject.

:Niranjan.
Logged

Ethan Hansen

  • Full Member
  • ***
  • Offline Offline
  • Posts: 148
    • Dry Creek Photo
Re: Optimizing printing profiles from i1Studio Spectrophotometer
« Reply #4 on: October 31, 2018, 04:30:13 pm »

](1) does using one of my own images to generate more patches and optimize the profile in any way make the profile less suitable for printing other images?[/b] I can imagine that if you 'push' the profile in one direction to make it better for one image, for other images the 'as pushed' profile may be less good than was the non-optimized profile. On the other hand, maybe adding any additional patches to what has already been printed, read, and incorporated can only make the as-optimized profile better for all images.

As Andrew mentioned above, letting i1Profiler automatically choose colors from an image appears to never make an appreciable difference. Selectively choosing color ramps for problem areas sometimes is successful. In our poking about with i1P performance, we found the profiles went off the rails if patch selection was not balanced enough. Cram too many patches of a particular color spread into i1Profiler and the resulting profiles do beautifully in that area but poorly overall. I suspect this behavior explains the results Niranjan saw as well.

And if that is the case, then (2) does performing successive optimizations using different images, at least up to a point, continue to improve the profile? I can imagine that it does, but I can also imagine that the software can only adjust through one set of optimization patches, in which case a second optimization would necessarily undo the results of the first optimization before incorporating the readings of the second optimizing patch set.

No benefit that I have seen. I tried just this experiment with an older version of i1P (ca. 2014-ish) and found that accuracy actually degraded with successive optimization rounds. Extracting the combined measurement data from the profile and simply building a new profile from that improved results.

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Optimizing printing profiles from i1Studio Spectrophotometer
« Reply #5 on: October 31, 2018, 05:25:23 pm »

As Andrew mentioned above, letting i1Profiler automatically choose colors from an image appears to never make an appreciable difference. Selectively choosing color ramps for problem areas sometimes is successful. In our poking about with i1P performance, we found the profiles went off the rails if patch selection was not balanced enough. Cram too many patches of a particular color spread into i1Profiler and the resulting profiles do beautifully in that area but poorly overall. I suspect this behavior explains the results Niranjan saw as well.

No benefit that I have seen. I tried just this experiment with an older version of i1P (ca. 2014-ish) and found that accuracy actually degraded with successive optimization rounds. Extracting the combined measurement data from the profile and simply building a new profile from that improved results.

This is what I have done since just optimizing with from an existing profile didn't preserve the sample set when I first tried their default approach. I think that may have been fixed but haven't checked it. I have also seen significant improvement from creating additional optimizing patches (not from an image however) with further improvement by including a larger number of near neutrals. But the latter is, I believe, something peculiar to the design of my 9800 which has a very lumpy neutral response.

One other thing is to save the optimized patch set then reload it. This gets rid of fractional RGB values made by the patch generator. Since the printed image is 8 bits this seems appropriate.
Logged

NAwlins_Contrarian

  • Full Member
  • ***
  • Offline Offline
  • Posts: 225
Re: Optimizing printing profiles from i1Studio Spectrophotometer
« Reply #6 on: October 31, 2018, 11:34:06 pm »

Ethan and Doug, thanks for the additional reports and suggestions. I ended up only creating one optimized profile, for Pictorico White Film in the Pro-100, but I kept the un-optimized profile too. I intend to make a pair of prints, one with each profile, of both the image used to optimize and another image that I think would look good on the Pictorico material but that has a very different set of colors than does the one I used for optimization.

Really, I think it would be nice if X-Rite provided software for the i1Studio (the successor to the ColorMunki Photo) that let you choose to use more than 100-150 patches for the initial profile, and more than 50 for optimization. But I guess that is part of what they use to encourage you to pay for to step up to a more expensive spectrophotometer kit.
Logged

felixpq

  • Newbie
  • *
  • Offline Offline
  • Posts: 24
Re: Optimizing printing profiles from i1Studio Spectrophotometer
« Reply #7 on: December 11, 2018, 12:18:08 pm »

I use ArgyllCMS tools to do my profiles, I can use as few or as many patches as I want. You can also create a device link profile that take into account the gamut of a single or a few images, your the boss. Using Argyll tools, I also make multiple reading of the printed patches, and average (median) them to reduce reading errors and less error in implies less error out or a better profile. But all this is at the cost of using cryptic command line tools mostly, though there are a few gui interface.

This as nothing to do with what the I1Studio software does with its optimisation thing. Argyll docs suggest from 500 to 3000 patches for a fair to excellent profile, so the I1Studio software must have an amazing algorithm to achieve good result with so few patches and it must also heavily rely on the cmm to make pretty good interpolation. With time and many optimizations, I imagine the 2 profiles could both produce excellent result on there own but if you add the image dependent gamut mapping Argyll propose, you should be ahead from the start.

I1Studio creates a CLut with a granularity of 33, that mean 35937 grid points. If anywhere in that space there is what I would call local nonlinearities, a low number of patches is more likely (statiscally) to miss that spot and any others that may be present. Yes one can argue that printers, ink and papers are made now with pretty good quality standards that this kind of local nonlinearities are practically non existant but what if you are the unlucky guy that bought an outlier?

Live long and prosper, or something like that.
YG



 
Logged
Pages: [1]   Go Up