Luminous Landscape Forum
Raw & Post Processing, Printing => Printing: Printers, Papers and Inks => Topic started by: Doug Gray on January 30, 2021, 08:28:50 pm
-
Being somewhat OCD in trying to get the highest quality printer profiles, I noticed an odd phenomena on my Pro1000. When printing colors tended to get darker towards the end of the printed page. The changes varied with colors. Oddly, the more saturated colors had relatively small changes but the darker grays had changes that were quite large. Differences were as much as 1 dE between colors within 2 inches of the bottom and the rest of the print.
So I made a repeating pattern of 4 grays, RGB (50,50,50), (100,100,100), (150,150,150), (200,200,200) on a standard, i1iSis US letter chart. This was measured and the measurements averaged for each of the 4 patch values. The average was then subtracted from the measurements and the result normalized to Lab=(50,0,0) to create a visual indication of spatial error. Here's the results:
(https://dm2305files.storage.live.com/y4mynYdvxleXEbXm3hWYa_6bB1h5uzeB2YGJxL0Lrr9fzdRD9Q4U7iphxVEIBtEDOe3lUDQ3dO0zhvU4G9GNQqCrtXoYQWobGI77l682LfDoAN6MCGM63ozxEv6zoEiElnt64GnQYgyRUKYDMUGh2cHQhwMAacjgldyO_3SZDE2w3s?width=1600&height=800&cropmode=none)
The left hand side is an image of what's printed. The two images to the right are the measured patch differences from the global averages for the Canon Pro1000 and the Epson 9800. For the Pro1000, the darkest patches RGB(50,50,50) and (100,100,100) get even darker by about 1dE towards the bottom of the image. This is highly visible because it's magnified 10x so shows as a dE of 10.
I tried feeding the paper in manually rather than from the top with the same results. My guess is that the difference is due to the paper bending slightly downward as it's near the end of the printing and that's causing a shift in paper to head spacing. There may also be an effect as the end of the paper clears the input feeding rollers.
My workaround is to limit US letter i1iSis charts to 27 rows instead of the default, 33 when printing charts for profiling.
-
If you are unhappy with that Canon behavior, run the same test on an Indigo digital press; you'll find a new love for that Canon. 😂
-
Thanks Doug. I have noticed similar behavior of inconsistency when making curves for QTR with my P800 forcing me to multiple patch sets on one A4 page with semi-random patch distribution. However, I have not measured how it behaves in different places of the page.
-
Thanks Doug. I have noticed similar behavior of inconsistency when making curves for QTR with my P800 forcing me to multiple patch sets on one A4 page with semi-random patch distribution. However, I have not measured how it behaves in different places of the page.
Hi,
I posted a process for creating B&W profiles for soft proofing and accurate Rel/Abs printing on the CM subforum and a link to the github rep for the source and executable. The default it uses is a 52 patch set with five duplications and is also randomized. Since it is well short of the page end the phenomena the P1000 has doesn't affect it. But multiple patches randomized is a good process best results. I see pretty consistent average dEs around 0.3.
-
(https://dm2305files.storage.live.com/y4mynYdvxleXEbXm3hWYa_6bB1h5uzeB2YGJxL0Lrr9fzdRD9Q4U7iphxVEIBtEDOe3lUDQ3dO0zhvU4G9GNQqCrtXoYQWobGI77l682LfDoAN6MCGM63ozxEv6zoEiElnt64GnQYgyRUKYDMUGh2cHQhwMAacjgldyO_3SZDE2w3s?width=1600&height=800&cropmode=none)
My workaround is to limit US letter i1iSis charts to 27 rows instead of the default, 33 when printing charts for profiling.
Doug, what is the paper transport direction in printing on the samples shown?
Any changes between higher and lower printing quality settings?
Met vriendelijke groet, Ernst Dinkla
https://www.pigment-print.com/spectralplots/spectrumviz_1.htm
"For every complex problem there is an answer that is clear, simple, and wrong." H. L. Mencken
-
Doug, what is the paper transport direction in printing on the samples shown?
Any changes between higher and lower printing quality settings?
Met vriendelijke groet, Ernst Dinkla
Hi Ernst,
Quality settings are at max except printing is bi-directional. The transport direction is top to bottom so the shift occurs near the last 20% of the page printing.
I haven't tested lower quality printing to see if the differences are larger or smaller.
Some time back I found the default settings on the 9800 produces similar shifts but along the vertical, changing from left to right. I was able to correct this by adjusting the vacuum level on the 9800 which now shows no positional differences.
I tried supporting the paper as it exited into the Pro1000 printer tray to reduce bending near the last of the printing but it made no significant difference.
-
My Epson 7900 shows similar behavior. I think this phenomenon is way more common than most people might think.
In my 7900 I have Piezography Pro inks loaded and I print with QTR. I first discovered this several years ago when linearizing a QTR quad file. (A quad file is not an ICC profile; it is more like a "media setting" file that controls how much ink and which inks are laid down for each pixel value [0-255] in the grayscale image being printed). The linearization process starts with a Piezography-provided grayscale target (see attached) that is printed then measured. The target is constructed in "column/row" order with patches from 255 to 0 running down column 1 first, then continuing in column 2 etc. I started noticing an occasional very odd repeating "ripple" in the measurements using my i1Pro2. The patches were scanned row by row. See the attached graph for an especially bad example of the measurement ripple. Most of the time it was not this bad, but still noticeable.
The data shows a bump/drop in density every 18 measurements. After a LOT of investigation I finally realized that the target had 18 rows (!) and ink density was trending higher in a non-linear fashion towards the end of the target when being printed which resulted in the measurement ripple when plotted from lowest L* to highest L*.
The actual cause of this is hard to pin down, but it could be any of several things (just my guesses here):
- print head heating up causing change in droplet characteristics resulting in non-linear change in ink density as page is printed
- movement of head, carriage and ink lines (sloshing) causes pigment density to change as page is printed
- change in ink line pressure during printing
- mechanical non-linearity in carriage movement/page feed that results in changed dither/dot placement as page is printed
In any case, it seems this behavior is more "normal" in large format printers than one would expect. It also seems to be worse with slower printing speed and high printing quality settings. When I printed the same target at a lower quality setting (720x1440 versus 2880x1440), the problem was completely eliminated. And I'm guessing that even within the same printer model, one machine may exhibit this more than another.
My solution going forward to control this to the extent possible was to start doing two things:
- Use scrambled (randomized) patch targets
- Use smaller patch sizes (but still meeting minimum requirement for whatever spectro is being used)
The smaller the overall target, the smaller the accumulating effect, and scrambled patches helps to average out the density drift.
And of course, this is one big reason why we use smoothing algorithms in making profiles (and in linearizing QTR quad files!).
Dave
Edit: Minor wording change to add clarity.
-
My Epson 7900 shows similar behavior. I think this phenomenon is way more common than most people might think.
In my 7900 I have Piezography Pro inks loaded and I print with QTR. I first discovered this several years ago when linearizing a QTR quad file. (A quad file is not an ICC profile; it is more like a "media setting" file that controls how much ink and which inks are laid down for each pixel value [0-255] in the grayscale image being printed). The linearization process starts with a Piezography-provided grayscale target (see attached) that is printed then measured. The target is constructed in "column/row" order with patches from 255 to 0 running down column 1 first, then continuing in column 2 etc. I started noticing an occasional very odd repeating "ripple" in the measurements using my i1Pro2. The patches were scanned row by row. See the attached graph for an especially bad example of the measurement ripple. Most of the time it was not this bad, but still noticeable.
The data shows a bump/drop in density every 18 measurements. After a LOT of investigation I finally realized that the target had 18 rows (!) and ink density was trending higher in a non-linear fashion towards the end of the target when being printed which resulted in the measurement ripple when plotted from lowest L* to highest L*.
The actual cause of this is hard to pin down, but it could be any of several things (just my guesses here):
- print head heating up causing change in droplet characteristics resulting in non-linear change in ink density as page is printed
- movement of head, carriage and ink lines (sloshing) causes pigment density to change as page is printed
- change in ink line pressure during printing
- mechanical non-linearity in carriage movement/page feed that results in changed dither/dot placement as page is printed
In any case, it seems this behavior is more "normal" in large format printers than one would expect. It also seems to be worse with slower printing speed and high printing quality settings. When I printed the same target at a lower quality setting (720x1440 versus 2880x1440), the problem was completely eliminated. And I'm guessing that even within the same printer model, one machine may exhibit this more than another.
My solution going forward to control this to the extent possible was to start doing two things:
- Use scrambled (randomized) patch targets
- Use smaller patch sizes (but still meeting minimum requirement for whatever spectro is being used)
The smaller the overall target, the smaller the accumulating effect, and scrambled patches helps to average out the density drift.
And of course, this is one big reason why we use smoothing algorithms in making profiles (and in linearizing QTR quad files!).
Dave
Edit: Minor wording change to add clarity.
Hi Dave,
Interesting. I did have positional shifts on my Epson 9800 until I adjusted the vacuum setting which then eliminated the problem. However, they were horizontal not vertical like my Pro1000 shows. Your ticks where the rows shift from bottom to top are consistent with the same thing I see. On the Pro1000 it's clearly a mechanical paper feed issue likely effecting slight changes in the paper/head spacing.
It's quite possible that a lower resolution would improve things if the ink droplets emitted are larger they should be less impacted by paper to head spacing. I'll check it on my printer.
As for thermal head heating, I did have that problem with the Canon 9500 II. It exhibited quite a bit of change when the same patch was repeated on the same horizontal line. I haven't seen that on the Pro1000 nor on the 9800.
-
In any case, it seems this behavior is more "normal" in large format printers than one would expect. It also seems to be worse with slower printing speed and high printing quality settings. When I printed the same target at a lower quality setting (720x1440 versus 2880x1440), the problem was completely eliminated. And I'm guessing that even within the same printer model, one machine may exhibit this more than another.
Dave
I just ran the test with quality set to high instead of highest. The effect decreased significantly. Largest repeatable change was .5 dE instead of 1 dE. And the effect was limited to the bottom 3 rows instead of the bottom 6 rows.
However, over the bulk of the middle of the paper the variation was a slightly higher but also appeared random as in the others.
Interesting.
-
I just ran the test with quality set to high instead of highest. The effect decreased significantly. Largest repeatable change was .5 dE instead of 1 dE.
These invisible and nearly invisible differences sure can be a bitch (if you let em). ;D
-
These invisible and nearly invisible differences sure can be a bitch (if you let em). ;D
They are, of course, while measureable, rarely visible differences. But when you are creating a colorimetric accurate target for calibrating other equipment it's good to remove as much variation as possible. As I said, I'm pretty OCD about this.
But I take your point. These errors are pretty small. Kind of like the difference in printing 8 bit RGB v 16 bit RGB. You really have to jump through hoops to see any effect.
-
They are, of course, while measureable, rarely visible differences. But when you are creating a colorimetric accurate target for calibrating other equipment it's good to remove as much variation as possible. As I said, I'm pretty OCD about this.
I agree on both points. ;) To a degree. Where to draw the line? 0.05 dE? 0.001 dE?
Have you run the same target say half a dozen times through the same iSis then compared the delta due JUST from the instrument? Or over the course of days? It too adds to the errors as I'm sure you know. And there isn't anything you can do about this invisible difference.
-
Interesting results from trying different configurations.
First, changing from top feed to back feed made no difference. Also adding a thick layer on top of the output tray to keep the printed paper from bending downward as it fed out made no difference.
However, changing from highest quality to high quality decreased the excursions from delta L* 1.1 to .7. Also changing from bi-directional to uni-directional reduced excursions from 1.1 to .9;
Here's the L* standard deviation of the mean of each row relative to the average of all 33 rows:
0.0650 Epson 9800, highest quality, bi-directional
0.1974 Pro 1000 high quality, bi-directional
0.3210 Pro 1000 highest quality, bi-directional
0.2456 Pro 1000 highest quality, uni-directional
Plots of the above:
(https://dm2305files.storage.live.com/y4mlG4bCk2af6BGxPvkvmpHlqQfE4spn8u8CnobElAeXjvb2k8sBf8uw7nUtfMVeT9Bv6T6kDT4ncU_TgdL61iWSxihcc4d82tHYnkh03laiRFC-9poRCaztsH0dkgQdyczE0gHy87e7eUmBxH3fxGPw12zLRpjwA8NH7AJlV-S7iY?width=770&height=656&cropmode=none)
-
Have you tried different papers? I'd be really interested in whether a 2x coated paper like the Moab Entrada 300 prints more evenly.
-
Have you tried different papers? I'd be really interested in whether a 2x coated paper like the Moab Entrada 300 prints more evenly.
The glossy's I've tried, from Platinum Pro to Costco, all have pretty similar shifts. However, matte, MP101 has a far smaller effect. Less than .3 shift end to end. Similar with Canson Photo Rag The Epson 9800 remains a big leader with far less variation on all of them. Now if only the 9800 didn't have the very visible problem of gloss differential and bronzing.
However, please note that these sorts of shifts are not visible on prints. Even the larger ones of 1.1 change. At least I can't see them. It's more something that becomes an issue when trying to create an optimal profile for lab work.
-
Hi, Doug: Would this effect go away on a roll paper or on a bigger size paper? Also, I thought Canon Pro1000 had a vacuum transport, isn't that not suffucient to take care of such problems? If not then, probably errors on something like Epson Surecolors might be even worse.
:Niranjan.
-
And there isn't anything you can do about this invisible difference.
I'm sure someone will be able to profile each part of the printer's output and then have multiple profiles, convert the chart to multiple converted versions and blend them together to keep their numbers right. ;-)
It's a hobby.
I prefer taking photos.
-
Hi, Doug: Would this effect go away on a roll paper or on a bigger size paper? Also, I thought Canon Pro1000 had a vacuum transport, isn't that not suffucient to take care of such problems? If not then, probably errors on something like Epson Surecolors might be even worse.
:Niranjan.
As far as roll paper goes, I can't say. I would expect less of an effect but only my 9800 has roll paper.
I wouldn't sweat these effects. They are really more interesting from a lab point of view. Because the shifts are gradual over about two inches you won't have any visual effects. 1 dE differences are only noticeable when two patches are very close to each other. Gradual changes are far less visible.
However, when comparing the quality of profiles made with really large numbers of patches, say 4000 or more, these effects significantly interfere with evaluating the resulting profiles because they introduce anomalies that are now baked into the profile. Again, the differences are not visible. It's more of just an experimental result. One gets perfectly good profiles with 1000 or more patches.
But if you are a bit OCD this effect will become significant with large patch counts. And one might well find a 6000 patch profile underperforming a 3000 patch profile purely as a result of where the patches happen to land on the paper.
For my purposes, I'm going to make patches with 27 rows instead of 33 on my iSis 2. It's about 20% more paper but the ink and scan time is the same.
-
I'm sure someone will be able to profile each part of the printer's output and then have multiple profiles, convert the chart to multiple converted versions and blend them together to keep their numbers right. ;-)
It's a hobby.
I prefer taking photos.
Might make a good grad student master's thesis at Rochester Institute of Technology. :)
Definitely an academic exercise.
-
Interesting.. I think it causes by platen height changes too, paper warp or bending.
Epson desktop printers manual points this out. In P600 manual shows "Top/reduced print quality area: 1.30 inches (33 mm) minimum" and "Bottom/reduced print quality area: 1.42 inches (36 mm) minimum". The numbers are the same with P700/P900. Actually, there has the same info in Canon PRO1000's Manual (attached image below). But Canon has a wider quality reduced area, top 53.3 mm, bottom 49.5 mm. It's roughly 2" in length.
Larger printers have vacuum suction keeping papers on the plate to avoid this issue. P7000, P9000, P7500, P9500 did not have that statement in manual. But in my own experience, P9500 has similar issue on top of the paper.
-
Interesting.. I think it causes by platen height changes too, paper warp or bending.
Epson desktop printers manual points this out. In P600 manual shows "Top/reduced print quality area: 1.30 inches (33 mm) minimum" and "Bottom/reduced print quality area: 1.42 inches (36 mm) minimum". The numbers are the same with P700/P900. Actually, there has the same info in Canon PRO1000's Manual (attached image below). But Canon has a wider quality reduced area, top 53.3 mm, bottom 49.5 mm. It's roughly 2" in length.
Larger printers have vacuum suction keeping papers on the plate to avoid this issue. P7000, P9000, P7500, P9500 did not have that statement in manual. But in my own experience, P9500 has similar issue on top of the paper.
Yep. Thanks for that. I never thought to look in the manual under specifications. I have 1.5" top and bottom margins. There's still some effects at 2" and have to go to 2.4" on the bottom and 2" on the top before the L* change drops below .3 That's not quite what the manual states but it's close.
So all this time people have been making profiles with significant portions of the printer area outside of the "good" zone. Wow.
As an experiment I also tried putting a gentle curve in the bottom of the paper in both directions. There was no difference and exactly the same effects occurred with both. So it isn't paper warp.
-
Yep. Thanks for that. I never thought to look in the manual under specifications. I have 1.5" top and bottom margins. There's still some effects at 2" and have to go to 2.4" on the bottom and 2" on the top before the L* change drops below .3 That's not quite what the manual states but it's close.
So all this time people have been making profiles with significant portions of the printer area outside of the "good" zone. Wow.
As an experiment I also tried putting a gentle curve in the bottom of the paper in both directions. There was no difference and exactly the same effects occurred with both. So it isn't paper warp.
Now we know. When all else fails read the instructions. ;)
-
Yep. Thanks for that. I never thought to look in the manual under specifications. I have 1.5" top and bottom margins. There's still some effects at 2" and have to go to 2.4" on the bottom and 2" on the top before the L* change drops below .3 That's not quite what the manual states but it's close.
So all this time people have been making profiles with significant portions of the printer area outside of the "good" zone. Wow.
As an experiment I also tried putting a gentle curve in the bottom of the paper in both directions. There was no difference and exactly the same effects occurred with both. So it isn't paper warp.
The double head printers of Epson, 75/95xx not excepted, all have a print quality issue at the beginning of a print. The driver has options to select how the printer should handle the first few inches (per edge quality).
-
In my case, (7900, see my above post), the phenomenon occurs regardless of where on the page the target is positioned. I don't believe it has anything to do with the "degraded print quality" that occurs in the 1" top and bottom margins of Epson printers. I ran many tests to verify this. So I am convinced that there is something else at play...and I lean towards something related to the ink delivery system up to and including the head.
-
In my case, (7900, see my above post), the phenomenon occurs regardless of where on the page the target is positioned. I don't believe it has anything to do with the "degraded print quality" that occurs in the 1" top and bottom margins of Epson printers. I ran many tests to verify this. So I am convinced that there is something else at play...and I lean towards something related to the ink delivery system up to and including the head.
With the Canon 9500II, there were significant history effects. That is the colors of adjacent patches impacted those nearby. For instance I would see a deltaE change of 1.5 on a dark patch if surrounded by dark patches. This doesn't occur on the 9800 or Pro1000. And on the 9800 there wasn't sensitivity to patches being printed near the top or bottom of the page but there was sensitivity to where the patches were located horizontally. This turned out to be fixable by tweaking the vacuum setting.
-
Holy cow. I took my optimized, 957 single page patch i1iSis chart which uses packed RGBs and 102 extra near neutrals and re-randomized it to place all the near neutrals away from the top/bottom edges so the most saturated colors are at the outer 10 rows. Then I measured all the device neutrals and the L* smoothness between steps improved from an rms value of .35 to .14. Since dE2000 is most sensitive near the neutral axis and drops in sensitivity by large amounts as colors become increasingly saturated, this suggests either profiles should be made with large margins (2.5") on top/bottom or patches should be arranged on the charts such that the most saturated colors are nearest the top/bottom.
This also explains why adding lots of near neutrals on large chart sets (>3k) made very small improvements in near neutral color accuracy. It may well be that the single page chart will now show better near neutral color accuracy than a 3k chart. That's next to investigate.
-
I did a personal test to figure out will this issue cause quality problem. A full letter size resolution test chart be made. CTF line pairs are between 20 to 800 dpi. Printed it, scanned it and analyzed it.
First attachment shows the image and analyzed area. Distance to the bottom edge of paper are 0.3 cm, 0.8 cm, 2.0 cm, 4.7 cm, 8.3 cm and 12.2 cm. I use ImageJ to analyze the contrast of the scanned file in these areas. Results are in second attachment. X axis is the position along CTF line pairs, Y axis is the gray value. The contrast in "image quality reduced area" is sightly lower. That means resolution near paper edge is lower than center of the paper.
It is tested on EPSON P600, would like to see any results on other printers.
-
Oh my. Turns out the patch location profiling the Pro1000 has really large effects on profile quality.
I've already discussed the impact of always printing profile targets after an initial page and discarding that page. This is critical even if you have already printed a page recently (within the last few minutes). There is some sort of change that occurs when there is any delay between printing successive pages. Here's an example of how bad it is.
The following is dE(1976) from 180 random colors in the printer's gamut and colors from the neutral axis from L*=3 to 93 in steps of 1.
Printing a 957 patch i1iSis page after the prior page was printed 8 hours earlier.
dE 180 color patches: 1.910 dE of 91 neutral patches: 1.089
However, discarding the first page and printing the chart immediately again reduces improves the profile accuracy.
The second 957 patch i1iSis page immediately after the prior page.
dE 180 color patches: 1.187 dE of 91 neutral patches: 0.570
And it gets better. Printing the same 957 patches on two consecutive pages where the page length was limited to 23 rows of 29 patches (667 total) improved the profile accuracy yet again.
dE 180 color patches: 0.819 dE of 91 neutral patches: 0.456
Summary: For the Pro1000, charts should be printed so the there is 2.5" margin along the long side of a US letter size (11" x 8.5") paper and an initial, first page duplicate discarded.
Notes:
1. Applies only to the Pro1000. My Epson 9800 does not exhibit significant near edge variations nor first page differences unless days unprinted have occurred.
2. This is for glossy prints at highest settings. Matte prints tend to have about 30% smaller errors.
3. The effect is really quite large. To the point that better profiles are made with 600 patches than 2000 patches when the first page is discarded and the charts provide 2.5" margins top and bottom.
-
Oh my. Turns out the patch location profiling the Pro1000 has really large effects on profile quality.
I've already discussed the impact of always printing profile targets after an initial page and discarding that page. This is critical even if you have already printed a page recently (within the last few minutes). There is some sort of change that occurs when there is any delay between printing successive pages. Here's an example of how bad it is.
The following is dE(1976) from 180 random colors in the printer's gamut and colors from the neutral axis from L*=3 to 93 in steps of 1.
Printing a 957 patch i1iSis page after the prior page was printed 8 hours earlier.
dE 180 color patches: 1.910 dE of 91 neutral patches: 1.089
However, discarding the first page and printing the chart immediately again reduces improves the profile accuracy.
The second 957 patch i1iSis page immediately after the prior page.
dE 180 color patches: 1.187 dE of 91 neutral patches: 0.570
And it gets better. Printing the same 957 patches on two consecutive pages where the page length was limited to 23 rows of 29 patches (667 total) improved the profile accuracy yet again.
dE 180 color patches: 0.819 dE of 91 neutral patches: 0.456
Summary: For the Pro1000, charts should be printed so the there is 2.5" margin along the long side of a US letter size (11" x 8.5") paper and an initial, first page duplicate discarded.
Notes:
1. Applies only to the Pro1000. My Epson 9800 does not exhibit significant near edge variations nor first page differences unless days unprinted have occurred.
2. This is for glossy prints at highest settings. Matte prints tend to have about 30% smaller errors.
3. The effect is really quite large. To the point that better profiles are made with 600 patches than 2000 patches when the first page is discarded and the charts provide 2.5" margins top and bottom.
Just wondering about the cheapest way to achieve the highest degree of accuracy (as I also have a Pro1000), what needs to be on the initial page which is discarded? As in, could the first page be just a few patches of each colour or perhaps some text, etc such that the wasted ink costs are minimised?
If not, does the entire chart have to be printed and discarded or would, perhaps, say 1/2 or 1/4 of the chart be sufficient to remove the variability?