Printer Gamma. Interesting differences between Epson 9800 and Pro-10/Canon 9500

[Doug Gray]

One of the more curious things I've noticed is the really large difference in the gamma of the rgb device drivers of these printers. My 9800 has a printer gamma of about 6 while most other printers I've checked their profiles for have a gamma more similar to standard RGB types or L* or between 2 and 3.

One consequence of this is that I need to have a much larger patch set when profiling the Epson 9800 than, say the Canon 9500 II. Effectively, the high gamma of the Epson 9800 means there is a far larger dE difference between adjacent patches with high RGB values than low RGB values. To get the same dE spacing at the high end of RGB values requires 4 to 8 times more patches on the 9800 than on the Pro-10 or 9500.

So I'm considering creating a patch set for the I1Profiler that is gamma scaled for the 9800. Thus, RGB(128,128,128) would become RGB(180,180,180), and RGB(63,63,63) would become RGB(128,128,128) while RGB(0,0,0) and RGB(255,255,255) would remain the same.

This printer gamma difference means the I1Profiler patch sets on my 9800 are far darker overall. The thing that kicked me into thinking more about this was noticing the additional patches added when improving a profile in I1Profiler were heavily skewed towards much brighter patches on the 9800 but not on the 9500. Anyone else notice this or know whether this would improve 9800 profiling and/or require much fewer patches for a specific level of accuracy?

[Tim Lookingbill]

Interesting issue, Doug.

I've often suspected over the years of use of various brands of imaging devices from scanners to displays to printers that the shape of their tone reproduction gamma curve was of main concern especially in reproducing tonal gradation below 128RGB mid gray and its slope toward black.

I don't think we really pay much attention to this aspect of judging quality. It's sort of like live adjustments of the EQ of our favorite music. We typically judge the overall sound or look (for images) as OK, but things appear off when we focus on specific areas of tonality.

There's no grayramp for sound but even judging quality of printers examining a printed grayramp just doesn't show how individual images are going to look in brightness overall unless one notes the separation of ramp patches coming out of black.

A 6 gamma doesn't seem like a normalized and reliable state to build profiles from measured patches.

I thought the digital age would finally settle on a Standard of image dynamics for all digital devices, but it appears from your post this isn't the case. I can't understand the "secret sauce" behind using 6 gamma with a printer. Unless it's a way to get the user to spend more on ink during the profiling process. Dark prints means more ink used.

In a related issue I also still can't understand how digital cinema can ruin the look of contrast of some movies which are suppose to adhere to a standard for digital projection. D-cinema standard is 2.6 gamma, but some movies botch the contrast so bad during projection it's unwatchable.

[Doug Gray]

Interesting issue, Doug.

I've often suspected over the years of use of various brands of imaging devices from scanners to displays to printers that the shape of their tone reproduction gamma curve was of main concern especially in reproducing tonal gradation below 128RGB mid gray and its slope toward black.

I don't think we really pay much attention to this aspect of judging quality. It's sort of like live adjustments of the EQ of our favorite music. We typically judge the overall sound or look (for images) as OK, but things appear off when we focus on specific areas of tonality.

There's no grayramp for sound but even judging quality of printers examining a printed grayramp just doesn't show how individual images are going to look in brightness overall unless one notes the separation of ramp patches coming out of black.

A 6 gamma doesn't seem like a normalized and reliable state to build profiles from measured patches.

I thought the digital age would finally settle on a Standard of image dynamics for all digital devices, but it appears from your post this isn't the case. I can't understand the "secret sauce" behind using 6 gamma with a printer. Unless it's a way to get the user to spend more on ink during the profiling process. Dark prints means more ink used.

In a related issue I also still can't understand how digital cinema can ruin the look of contrast of some movies which are suppose to adhere to a standard for digital projection. D-cinema standard is 2.6 gamma, but some movies botch the contrast so bad during projection it's unwatchable.

The specific gamma of the raw printer interface is pretty arbitrary since profiles, either in the driver itself, or in custom profiles will deal with it. At least within reason.

A gamma of 6 is pretty strong and requires more 3DLUTs to avoid larger rounding errors but it's not as awful as it might seem at first. Lab has a gamma of 3.0 with a large linear front end that drops it's effective gamma.  A gamma of 6 is really similar to a gamma of 2 on top of the Lab gamma which is what the 3DLUTs in a profile use.

I'm thinking of making profiles using the minimal patch set for speed of testing (283, I think) with and without a gamma correction then checking them against a different set of Lab patches to see how much of an improvement scaling makes. I certainly expect some improvement but the only way to know is to test it. If the difference is significant enough I'll check a larger set.

However, this may be an issue only in older printers and the 9800 is rather long in the tooth at this point. But the 98000 produces more repeatable colors from day to day than the 9500 which seems to be more sensitive to changes in temperature or humidity or something.

[GWGill]

One of the more curious things I've noticed is the really large difference in the gamma of the rgb device drivers of these printers. My 9800 has a printer gamma of about 6 while most other printers I've checked their profiles for have a gamma more similar to standard RGB types or L* or between 2 and 3.

That's more in the direction (or worse) of a raw inkjet response, where dot gain can be extreme, depending on the level of dot overlap.

So I'm considering creating a patch set for the I1Profiler that is gamma scaled for the 9800. Thus, RGB(128,128,128) would become RGB(180,180,180), and RGB(63,63,63) would become RGB(128,128,128) while RGB(0,0,0) and RGB(255,255,255) would remain the same.

A more standard approach is to use a separate calibration and profiling step.
(In ArgyllCMS I have a "gamma" value that can be imposed on test charts, and this is typically used when creating per channel calibration charts for raw inkjet output. With the calibration in place, the overal response is more perceptual, so a normal profiling chart is sufficient.)

Graeme Gill.

[Doug Gray]

That's more in the direction (or worse) of a raw inkjet response, where dot gain can be extreme, depending on the level of dot overlap.A more standard approach is to use a separate calibration and profiling step.

Yeah. It's surprising to see it on the RGB interface.

(In ArgyllCMS I have a "gamma" value that can be imposed on test charts, and this is typically used when creating per channel calibration charts for raw inkjet output. With the calibration in place, the overal response is more perceptual, so a normal profiling chart is sufficient.)
Graeme Gill.

That's certainly where I'd expect to see a gamma setting and something that is around Lab, a perceptual, gradient, would make sense. But I only have an RGB interface so the curiosity is whether I can improve profiles by re-mapping standard RGB patch sets. When printed on the 9800, the re-mapped sets do look much closer to the regular sets printed on the 9500. It may not improve things though since it really stretches the distance between the larger RGB values and I have no idea how linear those will be.

[Ernst Dinkla]

That's more in the direction (or worse) of a raw inkjet response, where dot gain can be extreme, depending on the level of dot overlap.A more standard approach is to use a separate calibration and profiling step.
(In ArgyllCMS I have a "gamma" value that can be imposed on test charts, and this is typically used when creating per channel calibration charts for raw inkjet output. With the calibration in place, the overal response is more perceptual, so a normal profiling chart is sufficient.)

Graeme Gill.

Could this be a reason: Printer resolution settings have a different effect on droplet sizes for today's piëzo and thermal head printers. Droplet size variation seems to be no longer a goal in thermal head designs, number of nozzles is however. The finest droplet size at the highest resolution of piëzo head printers is smaller than found in the highest resolution thermal head printers. The total dots boundary length that can create dotgain becomes longer with tiny dots than with bigger dots. Is there a gamma effect seen when lower resolution settings are used for target printing? As far as I know at the Epson highest resolution it only uses the smallest droplet available.

Despite being RGB-devices in driver + CM both HP and Canon wide format printers have better calibration features than Epsons have, assuming all used in RGB driven mode. Whether that is to compensate the supposed faster wear of the thermal heads is open for discussion but at least the calibration is integrated in Canon and HP wide format printer software/hardware. Calibration serves new profile creation and keeps existing profiles usable. That it allows lower patch numbers in targets is then another pro. Interesting as there is just a thread in the printers forum here on the topic of increasing the patch numbers of targets when using the HP Z3200.

In what way wear affects the output of piëzo heads is unknown to me, could be more or less ink squirted or no change. Literature states that there is no wear in piëzo heads, bad maintenance and unsuitable inks are mentioned as the real cause of issues. So the older 9800 should not lay down too much ink.

BTW, do the i1 Publisher added brighter patches improve the profile quality in that range significantly?

Edit: Epson Color Base calibration software aimed at the 9800 too.

Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
November 2016 update, 700+ inkjet media white spectral plots

[Mark D Segal]

.............Literature states that there is no wear in piëzo heads, bad maintenance and unsuitable inks are mentioned as the real cause of issues. ...............


Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
November 2016 update, 700+ inkjet media white spectral plots

Ernst, could you point us to the literature making this claim. I ask because I find it surprising based on the fact that people do report needing to replace these heads. Perhaps much depends on how one defines "bad maintenance", so it would be good to unpack that concept too.

[Ernst Dinkla]

To be more correct I did some googling on wear (and have done before) and did not see any mention of it in different articles. The one I quoted from is this one;

https://www.sgia.org/system/files/journal/14-jf-innovations-inkjet-printhead-technology.pdf


Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
November 2016 update, 700+ inkjet media white spectral plots

[Doug Gray]

Overall profile accuracy was reduced, not improved. Only the neutral axis showed improvement.

Conditions: The TC283 and gamma modified TC283 were used to create profiles. The same, 170 patch set of Lab color patches was then printed using each profile. The patches were then scanned and their Lab values compared to the reference image.

Average dE2k was 1.2 with the unmodified TC283 patch set and 1.6 with the modified set. Max was 3.7 and 4.9, respectively.

Interestingly, the set of neutral (gray) patches showed an improvement going from 1.1 to .6 dE2k.

Further investigation shows the relevance of Graeme's comments.  My initial motivation to do this was based on observing the differences between the Canon 9500 and Epson 9800 when identical patch sets were printed and noting the 9800 were much darker. Interestingly, the high gamma does not show up on the a* and b* dimensions but only on the L. Along the a* and b* (ramping device rgb along a hue angle and looking at the saturation), the 9800 with the unmodified chart shows much lower gammas.  As a result the modified chart produced a better match only along the neutral axis with worse match along the a* and b* plane.

[Mark D Segal]

To be more correct I did some googling on wear (and have done before) and did not see any mention of it in different articles. The one I quoted from is this one;

https://www.sgia.org/system/files/journal/14-jf-innovations-inkjet-printhead-technology.pdf


Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
November 2016 update, 700+ inkjet media white spectral plots

Thanks Ernst - interesting article. One respondent thinks one to two years is a reasonable expectation. Much probably depends on how it is used (for many people they will last much, much longer than that) - but we're wandering O/T I think.

[GWGill]

But I only have an RGB interface so the curiosity is whether I can improve profiles by re-mapping standard RGB patch sets.

Yes - should work, but per channel calibration curves (even in RGB) are better since they can be more
detailed, and make the cLUT do a lot less work, allowing it to be more accurate.