Strange Blue Gradient Profile in Perceptual Intent from I1Profiler

[Doug Gray]

A profile I use that has been flawless on colors in gamut has a most peculiar handling of saturated but real blue colors.

Specifically, when trying to print the color Lab(50, 0, -85) what is actually printed is Lab(33,-5,5) which is a darkish green far from the printer's gamut edge. All intents should print significantly OOG colors somewhere near the gamut boundary. In particular, Perceptual Intent should best render a smooth and pleasing transition from colors that are in gamut to those outside of gamut. This profile fails:

The three charts show hue shift from a gradient that goes from Lab(50,0,0) to hues evenly spaced at 10 degree intervals from 0 to 350 degrees.

Note the extreme deviation of the 270 degree hue on the left hand charge (Perceptual Intent) as saturation increases. It actually wraps around most of the left hand side of the a*,b* area before returning.

The gradients below each chart show how each intent renders the blue gradient. The perturbation in the Perceptual intent gradient rendering is underlined.

I would never normally encounter this sort of thing as I typically use Rel. Col. and almost always stay inside the gamut. But it might cause some strangeness for those using ProPhoto and bumping up saturation. This does show up on soft proofing which matches the print so at least the profile is reporting its errant behavior correctly.

[Doug Gray]

Lab constant L circles with Saturation going from 0 to 100.

One attachment I had to remove is in Lab 16 bit colorspace. I had to remove it as Lab tif files don't render here, the other is in ProPhoto 8 bit RGB showing how the circles are rendered when printing using Perceptual Intent with the buggy profile.

I've checked other profiles I have made and even made one using the same spectral data as the buggy one with PM5 and none of them exhibit this behavior.

Update: I've gone though all my profiles (about 50) and a few also show this bug including profiles for the Canon 9500 II. The problems are always in the same rough area and are only on Perceptual Intent.

I have also run I1Profiler using a single CPU core (to eliminate multi-core code bugs) on two computers, a Win 10 and a Win 7 with the same results.

Interestingly there is some variability changing the "smoothness" value when creating a profile.
Added Perceptual mapping for smoothness ==0, 20, and 60 (40 works)

May have just stumbled on a strange bug.

[Doug Gray]

Turns out it shows up in the Adobe RGB colorspace too. Here's a print with original, proof view and a cell phone snap with a particularly bad dark smudge on the perceptual intent.  Quite odd that nothing at all shows up in Rel Col, Sat, Col, or Abs Col. Only Perceptual and only on a small percentage.  Even there, few fall into the Adobe RGB colorspace.

[GWGill]

Note the extreme deviation of the 270 degree hue on the left hand charge (Perceptual Intent) as saturation increases. It actually wraps around most of the left hand side of the a*,b* area before returning.

Typically such behaviour is a numerical bug in the software. If you report it to X-Rite and give them the patch data, I imagine they will fix it.

[Doug Gray]

Typically such behaviour is a numerical bug in the software. If you report it to X-Rite and give them the patch data, I imagine they will fix it.

I can zip up the measurement files and Lab circle demos. I'll email them to a CSR.  I've been looking at it more closely. It is extremely sensitive to the specific luminance slice with big differences only 1 delta L apart. Sometimes turning white, other times a dark green in the same places 5 delta L's apart. I've found a few places it shows up inside sRGB though still about 20 dE outside the printer's gamut.

[Doug Gray]

Well, Here's an example of a profile made with I1Profier and Isis chart reader. The circles are different L* values with the hue rotated a full 360 degrees and saturation increasing from 0 to 120. There are grids shown at intervals of 10 in a* and b* to facilitate identifying the region where colors are messed up. For some reason I am seeing this effect more on M2 (uV cut) profiles than M0 ones from the same scans.

Note that the lab_circles.tif file is in LAB space which is a very large colorspace akin to ProPhoto though smaller is some areas. But it shows how profiles react to more extreme colors. All LAB colors are within the working space of ICC printer profiles. Caution, like ProPhoto, may of these LAB "colors" are imaginary. They exist as LAB numbers but are not actual colors in the real world. However, the colors they convert to and print are real.

For those interested, here's how to check your profiles to see how they map these colors.

1. Load the lab_circles.tif from the attached "strange.zip" file in Photoshop.
2. Select Perceptual Intent and click the "Preview" checkbox.
3. Select View->Proof Setup->Custom, the choose a printer profile in "Device to Simulate"

Now you can just scroll through your profiles and see what effect each of them has on these LAB circles. To see the effect of smaller working spaces just convert the lab_circles image to ProPhoto, Adobe RGB, or sRGB then run through the above process.

[samueljohnchia]

Hi Doug, check your PM. Thanks!

[unesco]

Doug, have you checked that all patches have been measured correctly, or printed correctly? I had similar problems with mistakes made during scanning (in greens) or some problems with print (blocked nozzles, some inconsistency of paper surface etc). There is even similar example (but in yellows) on Ilford web page for one of Epson 1400 profiles...

[Doug Gray]

Doug, have you checked that all patches have been measured correctly, or printed correctly? I had similar problems with mistakes made during scanning (in greens) or some problems with print (blocked nozzles, some inconsistency of paper surface etc). There is even similar example (but in yellows) on Ilford web page for one of Epson 1400 profiles...

The patches are fine. Under .5 average dE2k. Also, the problem only shows up on Perceptual. Relative, Saturation and Absolute all look and behave as expected. There is something about the way their Perceptual algorithms transition from in gamut to out of gamut that is producing this stuff. And it's in the BtoA0 tables too. It prints exactly like it looks. But then soft proofing uses AtoB1 in all modes.

Also, it is largely something that will only show up in Lab or ProPhoto RGB colorspaces where the colors involved are outside the printer gamut. More a curiosity for me as I rarely use Perceptual unless printing something off quickly.

Also checked the RGB patch values and none of them are 50 or 100 per a PM suggestion by Samuel.

[Ethan Hansen]

Doug - What target did you feed i1Profiler? I ask because we saw i1P create this type of artifacts when using customized targets with added patches that did not play well with the internals of i1Profiler. If these artifacts are from an i1Profiler target, X-Rite send the profile to X-Rite and they should be able to reproduce the problem.

One would hope X-Rite will investigate even if the targets are customized ones of your own design. I agree with Graeme that this points to a calculation bug in the Perceptual algorithm. I would not be surprised if the root cause is similar to a flaw we uncovered in our own perceptual and saturation rendering code 6 or 7 years ago. The behavior looks identical except that we ran into problems in out-of-gamut reds rather than greens.

[Doug Gray]

Doug - What target did you feed i1Profiler? I ask because we saw i1P create this type of artifacts when using customized targets with added patches that did not play well with the internals of i1Profiler. If these artifacts are from an i1Profiler target, X-Rite send the profile to X-Rite and they should be able to reproduce the problem.

One would hope X-Rite will investigate even if the targets are customized ones of your own design. I agree with Graeme that this points to a calculation bug in the Perceptual algorithm. I would not be surprised if the root cause is similar to a flaw we uncovered in our own perceptual and saturation rendering code 6 or 7 years ago. The behavior looks identical except that we ran into problems in out-of-gamut reds rather than greens.

Ethan,

I suspect the problem is due to ill conditioned matrixes in trying to create a perceptual mapping. Compression at the edges can generate this sort of thing. Rel Col. doesn't exhibit this at all but the criteria appears to just be minimizing dE from the gamut surface. Perhaps that makes for better conditioned matrixes. Just a wild guess though.

I first noticed this on profiles I made with I1Pro 2 where I'd added near neutral patches. However, the last image I posted uses an I1Profiler generated set of 2553 patches.

As bad as this looks, almost all the streaks are outside the printer's gamut and many are outside ProPhoto. A few of the worse just barely goes inside the printer's gamut.

The downloadable zip image is in Lab format with the saturation circle going up to 125. Each 10 a* and b* are delineated with tick marks in a grid so it's easy to locate the actual color Lab coordinates producing the problem. Also, each of the circles is a constant L* in steps of 5 from 5 to 95 so it's a pretty useful tool to check gamuts. If you use Photoshop's OOG mask it outlines the gamut at each L* slice quite nicely.

A note on Photoshop's OOG mask. It goes right to the boundary of RGB matrix profiles but overstates gamut by 6 dE for printer profiles.