Luminous Landscape Forum

Raw & Post Processing, Printing => Colour Management => Topic started by: Doug Gray on March 23, 2021, 09:28:26 pm

Title: Profiling a Canon Pro1000 is extremely sensitive to printing methodology.
Post by: Doug Gray on March 23, 2021, 09:28:26 pm
Almost all focus on profiling printers has been the number of patches and size of patches. This turns out to be wrong in that most of the error associated with Pro1000 profiles has little to do with these. On the Pro1000, they are secondary or even tertiary factors.

Over a year ago I noticed that the Pro1000 had a "warm up" problem and that if I printed 2 identical pages consecutively the second page had a pretty large color shift. So I got in the habit of always printing profile chart pages consecutively with a "throw-away" page at the start. This significantly improved profile color accuracy.

Then in January, I ran across another Pro1000 issue, color shifts based on the location on the printed page. See this:
https://forum.luminous-landscape.com/index.php?topic=137534.0

MfAlab pointed out that the Pro1000 (and some others) suggest limiting printing using an approx. 2" top and bottom margin for the most accurate colors.
https://forum.luminous-landscape.com/index.php?topic=137534.msg1206689#msg1206689

So to quantify how much of a factor this was, I ran three experiments using exactly the same set of 957 RGB patches which were from XRite's single page i1iSis chart called: "default-i1iSis."

Three profiles were made from the same 957 patches but the charts were printed differently. Then a set of 91 neutral patches from L*=3 to 93 and a set of 180 color patches randomly distributed in the printer's gamut were evaluated for each of the three profiles. The result was average dE 2000 for the 91 neutral patches and the 180 color patches for each set.

The differences in dE 2000 from profiles made with exactly the same patches were shocking.

Initial, one page 957 patch set: Colors: dE:1.184, Neutrals: dE:0.979
Second, one page 957 patch set: Colors: dE:0.661, Neutrals: dE:0.600
Two pages split with the 957 patch set and 2.5" top/bottom margins: Colors:0.450, Neutrals: 0:507

These changes are so significant that improvements by increasing patch counts are more than offset by errors printing patches using the full paper length. So much so that printing a patch set with a simple 6x6x6 grid and an extra 102 near neutrals for a total of 318 patches, using the optimal process with large, top/bottom margins has better performance than the 957 patches on a full page :
Colors dE:0.652, Neutrals dE:0.397

Here's the exact process for making the 3 profiles each from the same 957 RGB patches.

1. Print the single page, 957 patch image twice without printing anything previously within the last hour except a verified nozzle check. This has 33 rows of 29, 6mm x 6mm patches.

2. Before the print queue finishes, print the 957 patch set divided up into 2 pages with 2.5" margin or more on the top and bottom of the US Letter page.

Then print and measure

* Important Note: This is specific to the Pro1000 and possibly other printers that mention increased top/bottom margin for best color. My Epson 9800 does not exhibit this effect.
Title: Re: Profiling a Canon Pro1000 is extremely sensitive to printing methodology.
Post by: Doug Gray on March 24, 2021, 12:33:10 am
Attached is a graph of 3 pages showing the average dE2000 of rows of 29 colors. Each row is the same set but scrambled so they are different columns. The plots are groups of 10 consecutive rows starting at the top and going to the bottom thus making 23 groups. The three lines are from 3 different prints.

This shows that the colors in the rows near the top or bottom have almost twice the average deltaEs.
Title: Re: Profiling a Canon Pro1000 is extremely sensitive to printing methodology.
Post by: Rhossydd on March 24, 2021, 03:58:26 am
The differences in dE 2000 from profiles made with exactly the same patches were shocking.
Shocking ???? not really, anything less than a 1dE is invisible.

As dE is a value of comparison you really ought to say what what you're comparing the print outs with.

It's also worth pointing out that any effect you're measuring here maybe just be a fault on one particular example and may not be seen on other examples.

Title: Re: Profiling a Canon Pro1000 is extremely sensitive to printing methodology.
Post by: Doug Gray on March 27, 2021, 12:04:27 am
Shocking ???? not really, anything less than a 1dE is invisible.

Sure dEs less than 1 are rarely visible but those numbers are averages. About 5% of the 180 color patches are more than twice the average. But what shocked me was that a 318 patch set, printed in the middle, would outperform a full page 957 patch set.

Specifically, the XRite default i1isis 957 patch set is composed of a 9x10x10 grid and 57 additional, near neutral patches. The 318 patch set that outperforms it has a 6x6x6 grid and 102 additional, near neutrals. Since the 318 patch set has more near neutrals the better performance of neutrals was expected and did occur. That the color patch accuracy of the 216 grid profile was as good as the 900 grid profile simply because it was printed in the middle of the paper is what shocked me.
Quote

As dE is a value of comparison you really ought to say what what you're comparing the print outs with.

To clarify, the dE's are the average dE's from 180 L*a*b* measured RGB colors and 91 measured neutrals v. the L*a*b* color image values which create the same RGB sets through the profile under test.
Quote

It's also worth pointing out that any effect you're measuring here maybe just be a fault on one particular example and may not be seen on other examples.
Quite possibly it varies printer to printer but it's quite repeatable on my printer. Further, as pointed out by the manufacturer, color accuracy degrades within 2 inches of the top and bottom. So it's somewhat likely the effect applies generally to Pro1000 printers. As I noted. I do not see this effect on the Epson 9800 and benefits more from larger patch counts.