Ad
Ad
Ad
Pages: [1] 2 »   Bottom of Page
Print
Author Topic: ColorMunki: Merit in Using Higher Patch-count?  (Read 8250 times)
BruceHouston
Sr. Member
****
Offline Offline

Posts: 308



« on: January 08, 2010, 09:33:28 PM »
ReplyReply

The ColorMunki test patch count for printer profiling is small compared to other systems.  X-Rite has provided the capability of increasing the patch count by reading a sample image and then producing additional test targets based upon color values in the sample image.  The latter procedure will presumably increase accuracy in the color/tonal ranges associated with the sample image, and is thus a means of "tweaking" a profile for a particular image, type of image, skin tone, etc.

Has anyone experimented with using this capability to increase the patch count generally when creating a printer profile?  That is, has anyone come up with one or more sets of patch images that could be loaded into ColorMunki using the "Load Image" function to produce addititional printed test patches to be scanned to improve the profile accuracy across the gamut?  Or is a point of diminishing returns reached after the two-level patch scanning using the standard procedure?

If a third or subsequent levels of patch printing/scanning could be useful in the generalized case, what patch images should be input to the Munki?  Are the spectral characteristics of ColorChecker sufficiently special to make it worthwhile to capture a Colorchecker image for this purpose?  (This is probably an over-simplification, given that presumably X-Rite would have included ColorChecker patches in the standard routine if there were merit in doing so.)

Any thoughts are appreciated.

Thanks,
Bruce
Logged
probep
Full Member
***
Offline Offline

Posts: 149


« Reply #1 on: January 09, 2010, 08:42:32 AM »
ReplyReply

Quote from: BruceHouston
The ColorMunki test patch count for printer profiling is small compared to other systems.  X-Rite has provided the capability of increasing the patch count by reading a sample image and then producing additional test targets based upon color values in the sample image.  The latter procedure will presumably increase accuracy in the color/tonal ranges associated with the sample image, and is thus a means of "tweaking" a profile for a particular image, type of image, skin tone, etc.

Has anyone experimented with using this capability to increase the patch count generally when creating a printer profile?  That is, has anyone come up with one or more sets of patch images that could be loaded into ColorMunki using the "Load Image" function to produce addititional printed test patches to be scanned to improve the profile accuracy across the gamut?  Or is a point of diminishing returns reached after the two-level patch scanning using the standard procedure?

If a third or subsequent levels of patch printing/scanning could be useful in the generalized case, what patch images should be input to the Munki?  Are the spectral characteristics of ColorChecker sufficiently special to make it worthwhile to capture a Colorchecker image for this purpose?  (This is probably an over-simplification, given that presumably X-Rite would have included ColorChecker patches in the standard routine if there were merit in doing so.)

Any thoughts are appreciated.
I've tested a ColorMunki Photo with additional patch lists generated from the images that are supplied with the ColorMunki software. For profiling I've used a Canon Pro9500 with Epson Archival Matte Paper.
ColorMunki printer profile is good enought even without additional patches. But additional patches do increase accuracy.
See the figure (L*=50) below:
- White line - original ColorMunki printer profile (100 patches)
- Red line - ColorMunki printer profile with additional patches (450 patches)
- green line - (for comparison) printer profile generated with i1Pro UV-cut and ProfileMaker 5.0.8 (2660 patches).
Logged
Alan Goldhammer
Sr. Member
****
Offline Offline

Posts: 1616


WWW
« Reply #2 on: January 09, 2010, 08:48:45 AM »
ReplyReply

I have used ColorMunki successfully to create profiles for my Epson 2880.  It does a very good job and I routinely use one of my landscape images to refine the profile.  I was able to compare a ColorMunki profile with one created with more sophisticated tools as I printed the targets that Museo used to prepare the 2880 profiles that are on their website.  My ColorMunki profile is virtually spot on with the Museo MAX profile that they did using three sheets of color patches.  I used the evaluation image that Jack Flesher prepared that can be found here.  One thing that I have noticed that ColorMunki users need to be aware of is that the initial profiling does not do a good job on the dark blacks as there is virtually no difference between patches 0-12 on the B/W scale.  This is not the case with the Museo prepared profile and one needs to refine the ColorMunki profile using an appropriate B/W print.  For my purposes this is not terribly crucial as I use the ABW profile that Eric Chan has prepared for me for all my B/W printing.

Hope this helps others as I find the ColorMunki a wonderful buy for the money.

Alan
Logged

BruceHouston
Sr. Member
****
Offline Offline

Posts: 308



« Reply #3 on: January 10, 2010, 01:35:26 AM »
ReplyReply

Thank you both; very helpful.

Best,
Bruce
Logged
BruceHouston
Sr. Member
****
Offline Offline

Posts: 308



« Reply #4 on: January 10, 2010, 02:55:54 AM »
ReplyReply

Quote from: probep
I've tested a ColorMunki Photo with additional patch lists generated from the images that are supplied with the ColorMunki software. For profiling I've used a Canon Pro9500 with Epson Archival Matte Paper.
ColorMunki printer profile is good enought even without additional patches. But additional patches do increase accuracy.
See the figure (L*=50) below:
- White line - original ColorMunki printer profile (100 patches)
- Red line - ColorMunki printer profile with additional patches (450 patches)
- green line - (for comparison) printer profile generated with i1Pro UV-cut and ProfileMaker 5.0.8 (2660 patches).

Probep,

It appears that building out the profile with additional patches using the Munki (the red outline of your gamut chart) also reduced the gamut somewhat in the reds.  How is this possible?  Does this imply that the profile enhancing algorithm can overwrite profile values that were previously written into the profile from previously-analyzed patches?  This seems like a strange result, particularly if such a large difference were to be caused by lack of repeatability in the process of adding one or more color values to the profile multiple times.  I would expect the profile gamut to get larger or remain the same when revised with additional patches, but never to get smaller anywhere.

Any ideas?
« Last Edit: January 10, 2010, 02:58:36 AM by BruceHouston » Logged
probep
Full Member
***
Offline Offline

Posts: 149


« Reply #5 on: January 10, 2010, 04:55:40 AM »
ReplyReply

BruceHouston
See two additional figures (for L*=36 and L*=81) below. Maybe they are useful for an analysis.
Again,
- White line - original ColorMunki printer profile (100 patches)
- Red line - ColorMunki printer profile with additional patches (450 patches)
- green line - (for comparison) printer profile generated with i1Pro UV-cut and ProfileMaker 5.0.8 (2660 patches).
Printer: Canon PIXMA Pro9500
Paper: Epson Archival Matte paper
Logged
probep
Full Member
***
Offline Offline

Posts: 149


« Reply #6 on: January 10, 2010, 06:17:48 AM »
ReplyReply

BruceHouston
BTW about accuracy of a ColorMunki spectro.
First figure shows color differences between original ColorMunki profile and the reference profile (generated with i1Pro UV-cut and ProfileMaker 5.0.8 for 2660 patches) for Lightness L(LHS)=0.5
 max delta E[ab]=10

Second figure shows color differences between ColorMunki profile with additional patches (450 patches in total) and the reference profile for Lightness L(LHS)=0.5
 max delta E[ab]=4
« Last Edit: January 10, 2010, 06:31:40 AM by probep » Logged
BruceHouston
Sr. Member
****
Offline Offline

Posts: 308



« Reply #7 on: January 10, 2010, 10:44:51 PM »
ReplyReply

Thank you Probep; your measurements really are fascinating.

Could it be generalized from these measurements that a standard Munki-generated profile exaggerates the gamut; that a Munki-generated profile refined with additional patches brings the gamut into a more conservative (and accurate) range; but that even the refined Munki profile exaggerates the gamut somewhat relative to the reference, the latter being the most conservative (and presumably the most accurate)?

This seems counter-intuitive to me.  I would expect profiling software to be more conservative (narrower gamut) with a lower-accuracy spectro/measurement sequence  and more liberal (wider gamut) with a higher-accuracy spectro/measurement sequence.  (After all, an incorrect direct lookup of a color that is out of the absolute range of the printer capability would presumably bypass the rendering intent logic and could thus cause strange results.)

It would be interesting to see if other printer/paper combinations show the same effect.

Could Andrew Rodney and/or other color gurus comment about what is going on here?

Thank you,
Bruce
« Last Edit: January 10, 2010, 11:23:26 PM by BruceHouston » Logged
Neuffy
Newbie
*
Offline Offline

Posts: 42


« Reply #8 on: January 11, 2010, 10:36:41 PM »
ReplyReply

I can't say this with enthusiastic declaration, but I've been taught that larger patch targets (eg. with my i1 Pro setup) result in more accurate profiles that have slightly smaller gamut.

I've always just assuming this to be the case, pending my finding time to examine this problem directly.
Logged

BruceHouston
Sr. Member
****
Offline Offline

Posts: 308



« Reply #9 on: January 12, 2010, 10:07:37 AM »
ReplyReply

Quote from: Neuffy
I can't say this with enthusiastic declaration, but I've been taught that larger patch targets (eg. with my i1 Pro setup) result in more accurate profiles that have slightly smaller gamut.

I've always just assuming this to be the case, pending my finding time to examine this problem directly.


That appears to be the case from the data above; I just wonder why.
Logged
Neuffy
Newbie
*
Offline Offline

Posts: 42


« Reply #10 on: January 12, 2010, 10:23:42 AM »
ReplyReply

Quote from: BruceHouston
That appears to be the case from the data above; I just wonder why.

I'd expect it's because non-linearity is now detectable along the edges of the gamut, and it can now properly reduce gamut to improve gamut-edge linearity. With fewer patches, it never sees this non-linearity, so there is no reduction.

In theory, you would get the best gamut using something like a 3-patch-set-per-ink, I think. No ink/Half ink/Full Ink. It'd be a very tiny patch set, but it should give you ink right up to the ink limit settings.
Logged

BruceHouston
Sr. Member
****
Offline Offline

Posts: 308



« Reply #11 on: January 12, 2010, 10:35:30 AM »
ReplyReply

Quote from: Neuffy
I'd expect it's because non-linearity is now detectable along the edges of the gamut, and it can now properly reduce gamut to improve gamut-edge linearity. With fewer patches, it never sees this non-linearity, so there is no reduction.

In theory, you would get the best gamut using something like a 3-patch-set-per-ink, I think. No ink/Half ink/Full Ink. It'd be a very tiny patch set, but it should give you ink right up to the ink limit settings.


That would make sense; thank you Neuffy.
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 8782



WWW
« Reply #12 on: January 12, 2010, 11:22:33 AM »
ReplyReply

Quote from: BruceHouston
Could Andrew Rodney and/or other color gurus comment about what is going on here?

First off, I don’t see the gamut size (bigger or smaller after iteration) being any kind of indicator of “accuracy” or quality.
Second, the idea as I understand it is, you add more patches to this iteration process to improve areas in the profile you initially might have seen. An example is less than ideal gray balance. Add a set of new gray patches and iterate and build a new profile. Hopefully gray balance is better.
The iteration is used for the initial profile by first reading the 50 patches and building 50 unique patches from this initial data. 100 patches isn’t many! I’ve found the product builds quite good profiles with only 100 patches to the somewhat well behaved devices I’ve tested (mostly Epson’s). They rival those built using a much more expensive Spectrophotometer (iSis) and ProfileMaker Pro with 1700 odd patches.

In theory, more patches are better. In a prefect world, we’d measure 16.7 million color patches. Not only what that take forever, the resulting profile would be larger than many of our images. So every profile building profile like this has to extrapolate from less data to map into the resulting color space. 1500 patches isn’t half as good as 3000 and so on, there’s a point of diminishing return.

So in terms of the Munki, I’d build the 100 patch profile and print a lot of work to evaluate its output qualities. Note this alone tells us nothing about the other tables, those used for previews (soft proofing) but that’s another post. If you see certain areas of color space that don’t look good to you, the idea is to try to build those colors into a set of new patches for further iteration of the profile.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
BruceHouston
Sr. Member
****
Offline Offline

Posts: 308



« Reply #13 on: January 12, 2010, 12:42:28 PM »
ReplyReply

[quote name='digitaldog' post='339505' date='Jan 12 2010, 11:22 AM']

"First off, I don't see the gamut size (bigger or smaller after iteration) being any kind of indicator of accuracy or quality."

Agreed.  Gamut size alone would not be an indication of accuracy or quality.  The question has to do with what appears to be a pattern from Probep's measurements, only one parameter of which is gamut size.  Assuming that his  i1Pro UV-cut/ProfileMaker 5.0.8 measurement system is more accurate than a ColorMunki system using any number of patches, and further assuming that a Munki-generated profile from more patchs is more accurate than one generated from fewer patches, is there an explanation for why the "more accurate" profiles tend to cover a smaller gamut than the "less accurate" profiles?  (For purposes of this discussion, let us define "accurate" as a condition whereby a set of data points measured with a "more accurate" spectro/software measurement system is repeatably closer in value (clustered more proximate) to a similar set obtained using a known high-accuracy (lab standard) instrument than the same set of data points captured with a "less accurate" spectro/software measurement system.


"Second, the idea as I understand it is, you add more patches to this iteration process to improve areas in the profile you initially might have seen."

Yes; this is understood as the "best practices" methodology.  The question is more of a theoretical one relating to what appears to be a pattern from Probep's measurements.
« Last Edit: January 12, 2010, 12:45:45 PM by BruceHouston » Logged
teddillard
Sr. Member
****
Offline Offline

Posts: 660


WWW
« Reply #14 on: January 24, 2010, 10:21:17 AM »
ReplyReply

FWIW, I just ran some tests on the Munki vs. the i1Pro, and was pretty impressed...  especially in light of my first experience with v1.0.  

I go into a little more detail in my h2h blog, but here's the ColorThink gamut picture... a far cry from before.  As far as if it's worth it, for the price, with the additional SW and stuff, if you have the money it seems worthwhile.  

...and, not intending to denigrate the color science being discussed here, but my bottom line is, "do the prints look good?  better than the canned profiles?  as good maybe as the more expensive device?"  ...that answer would be a "yes".



Here's the blog post:
http://www.h2hreviews.com/blog/Revenge-of-...ear-later-.html
« Last Edit: January 24, 2010, 12:24:07 PM by teddillard » Logged

Ted Dillard
probep
Full Member
***
Offline Offline

Posts: 149


« Reply #15 on: January 24, 2010, 10:59:52 AM »
ReplyReply

Once more again about the accuracy of printer profiles.
I have evaluated the accuracy of profiles using (printing and measuring) the "adapted" Ugra/Fogra Media Wedge CMYK v3.0 testchart.
Printer: Canon PIXMA Pro9500
Paper: Epson Archival Matte Paper
1. The ColorMunki printer profile, standard profiling (50+50 patches)
   avr dE00=1.6, max dE00=3.3
2. The ColorMunki printer profile, profiling with additional patches (450 patches in total)
   avr dE00=1.52, max dE00=3.11
3. The printer profile created with i1Pro UV-cut and ProfileMaker 5.0.8 from 283 unique patches
   avr dE00=1.81, max dE00=3.87
4. The printer profile created with i1Pro UV-cut and ProfileMaker 5.0.8, TC9.18 (918 unique patches)
   avr dE00=1.08, max dE00=2.85
5. The printer profile created with i1Pro UV-cut and ProfileMaker 5.0.8 from 2244 unique patches
   avr dE00=0.98, max dE00=2.81
« Last Edit: January 24, 2010, 11:03:39 AM by probep » Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 8782



WWW
« Reply #16 on: January 24, 2010, 11:45:37 AM »
ReplyReply

Quote from: probep
Once more again about the accuracy of printer profiles.
I have evaluated the accuracy of profiles using (printing and measuring) the "adapted" Ugra/Fogra Media Wedge CMYK v3.0 testchart.

Can you explain the steps used to produce these values? TIA.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
Czornyj
Sr. Member
****
Offline Offline

Posts: 1374



WWW
« Reply #17 on: January 24, 2010, 12:11:13 PM »
ReplyReply

Quote from: probep
Once more again about the accuracy of printer profiles.
I have evaluated the accuracy of profiles using (printing and measuring) the "adapted" Ugra/Fogra Media Wedge CMYK v3.0 testchart.

I think that Fogra MW3.0 is intended to control the stability of your printer and paper rather than for profile quality check.
« Last Edit: January 24, 2010, 12:11:53 PM by Czornyj » Logged

digitaldog
Sr. Member
****
Offline Offline

Posts: 8782



WWW
« Reply #18 on: January 24, 2010, 12:16:08 PM »
ReplyReply

Quote from: Czornyj
I think that Fogra MW3.0 is intended to control the stability of your printer and paper rather than for profile quality check.
Agreed. In fact, the colors used for these evaluations IMHO require a custom built reference.

I can explain how I would go about the process. I’d use ColorThink for one, its invaluable for this kind of work for creating custom targets for profile evaluation, not profile creation. You can do this with a combo of Photoshop of course, ColorThink and ColorPort from X-rite (free).

First, the idea is to examine the expected values for patch colors and those that you measure through the profile. A target is necessary of course. First thing is deciding what patches are appropriate and hence, the reason it makes some sense to build a custom target instead of using the target that built the profile. For example, lets say you work in Adobe RGB (1998) as your working space. It makes little sense to test your profile using a target where some of the patches fall outside Adobe RGB (1998). You’re asking for errors. So building a target in that color space with “legal” values is helpful while the target you used to build the profile probably have out of gamut colors. The same is true when building CMYK profiles. The targets are all 400% TAC which you’ll never run. Using such a target for these tests skew the actual reported data you get.

Next you can save off this file such that every pixel is a single color (using Nearest Neighbor will do this). If your target has 33x45 rows and columns, you’ll need a 33x45 pixel document to feed to ColorThink and ColorPort. Prior to any conversions, this would be the “source” (reference) data. Save that off as a TIFF. Next, convert the original TIFF from (in this case, Adobe RGB (1998) ) through the profile you’re checking. You would want to do this with Perceptual and Colorimetric unless you have a workflow where the rendering intent is fixed, then just use that. You want to do this in Photoshop so you can use Black Point Compensation. You could do this conversion in ColorThink using color lists but it doesn’t yet support BPC. Save this converted document (lets say for simplicity, you just converted using Perceptual intent). You now have two small TIFFs where each pixel is one color in your target. One is the reference, one is the result though the profile. Expected and resulting Lab values.

If you load one of the Tiffs into ColorThink, you can build a ColorList using each individual color. Save it out as Lab data as a txt file. Do the same with the other TIFF converted with your profile from Photoshop. You now have two text files in Lab, one is the reference (colors in the target), the other are the results from the profile. You drag them into ColorThink and click on the deltaE button and there you go, average, max, etc values and you can sort by highest deltaE to see where in colorspace, you’re having issues.

To create custom targets to print and measure, you can use ColorPort. It will import a color list, build off targets (Tiffs that will print and read through a supported instrument) and measure the targets too if you wish. Be careful with UV Cut and non cut measurements. Ideally you’d do this test both ways. OBAs can really skew things (fun finding a paper with a -6 b*). This is useful because you don’t have to rely on the targets from Munki which are far smaller and different than the targets from ProfileMaker or whatever product you used. And since its the same target you built for both profiles, its apples to apples comparison.

Then the question becomes, depending on the device, not only how large was the target used to BUILD the profile, did it get printed a number of times in different patch sizes and orientation, then averaged (with outliers being tossed)? For presses, that’s super important. For our ink jet printers, maybe overkill but might be useful. And again, the patch size used for building the profile is pretty important for Spectrophotometer’s that average measurements per patch as an EyeOne Pro or iSis can. For example, for digital press profiles I’m building, I have one ECI target that has patches so large, it takes 3 11x17 patches to print out just this one target. Then its rotated (three more pages) and there’s a two patch, smaller patch target also output in both orientations. Total four ECI targets (10 pages) all averaged for a single press (now do this with each press and average all that data to build the profile).
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
probep
Full Member
***
Offline Offline

Posts: 149


« Reply #19 on: January 24, 2010, 12:27:39 PM »
ReplyReply

Quote from: digitaldog
Can you explain the steps used to produce these values? TIA.
First of all I do not know any common used method to evaluate the accuracy of an "RGB-"printer, http://luminous-landscape.com/forum/index....showtopic=41048
Moreover the patches from ColorChecker SG are out of the "printer-matte paper" gamut.
Therefore I got the chart from Ugra/Fogra Media Wedge CMYK v3.0 (72 patches) and the measured (reference) LAB values from http://www.fogra.org
I generated the TIFF chart from the reference LAB values using the ColorLab utility. This TIFF file was created in LAB workspace.
Then I printed TIFF charts with Absolute Colorimetric RI in Photoshop using each of 5 printer profiles.
Then I measured patches of the printed charts with i1Pro UV-cut.
At last I compared measured LAB values with the reference.
« Last Edit: January 24, 2010, 12:30:25 PM by probep » Logged
Pages: [1] 2 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad