Profiling printer, with scanner and Wolf's IT8. what am I doing wrong?

[Hale_JP]

I am using cheap Chinese inks, and decided to profile my printer for such particular conditions.

I got Wolf Faust's Kodak R1 target.
The hardware is simple,
Epson V33 (aka S630) scanner,
Epson PX-204 4-color printer
Fujitsu WPHS PRO paper.

The results I get is tanned skin goes pinky, greenish ocean water goes sky-blue. Like too little green. But I wouldn't say printer saves Y&C inks... it drinks them by galons.


If I tell you the process in Argyll I use, maybe you can find a flaw?

1. I scan IT8 with a black photo-envelope at the back (the lid is white) in VueScan RAW TIFF mode. Clean the dust in photoshop or GIMP without adding a profile.

2. Read values with Argyll and create a profile in "colorimeter"-mode with "absolute" intention. High quality regime. Ultra fails for some reason with "out of gamut WP".

>> following is a calibration+profiling sequence for the printer >>

3. Print one page test pattern, created with "-p 1.2" bias for dot-gain. (240 with something patches per page in Argyll, if I am not mistaken)

**I print it from MS Paint, since Photoshop keeps transforming colors, even when "printer manages" is selected and "no management" is in Epson driver settings. (Photoshop add "proofing" like paper tint, while white in Paint it is just zero-ink on paper in "no management" driver mode)

4. Scan it raw and read with Argyll, taking into account the scanner profile created in step-2.
5. Make a .cal calibration, and build .mpp model for the printer.
6. print a profiling target, taking into account the model and calibration from step-5.
6. Scan it, read it with the same scanner profile fed to Argyll.
7. make a temporary profile "colprof -v -qm -b -cmt -dpp Prn_prof"
8. Since I did not get how to use -kz and -kx options, I just try making all curves appear monotonously growing at least up to 75% and not disappear (like M disappears with defaults)
It may be like "xicclu -g -kp .005 10 .98 .78 0.5 -l260 -fif -ir PrinterBt.icm
-l260 is just a guess, since above 260 the curves do not grow, or distort.
I didn't make a screenshot, but it was approximately like on the hand-drawn image below.
9. I create ICC for .cal, ICC for the profile with sRGB.icm provided as a source space, and -kp with K-curve above.
10. Link them collink -v -qm -s -ip -op cal.icm profile.icm EpsonPX240.icm

Print, and get tints a bit wrong. Particularly fresh pink skin where it should be sun-tanned is a problem.

[smthopr]

I can't follow your workflow, and I'm not so good with command lines to use Argyll...

But, I've personally never had much success with scanner based measurements to create printer profiles. This was using commercial software "designed" to make printer profiles using a flat bed scanner.

Since you seem to be comfortable with Argyll, why not just buy a used iOne Spectro (you won't need a software license) for cheap and make your profiles this way?  Even a slightly out of calibration spectro should give better results than a flat bed scanner and .it-8 target.

[digitaldog]

But, I've personally never had much success with scanner based measurements to create printer profiles.

Ditto and I've tried a number of such products over the years. Always awful. Because a scanner is absolutely the wrong tool for this task. A Spectrophotometer is.
 

If I tell you the process in Argyll I use, maybe you can find a flaw?

Wrong tool for the job!

[Doug Gray]

Using a scanner to create a printer profile is an exercise in futility.

Scanners do not exhibit good Luther/Ives characteristics hence can't colorimetrically scan objects of different spectral reflectance characteristics. Profiling to an IT8 target gets a profile that is good only for scanning media using the same colorants as the IT8 target but even that doesn't address the next problem.

In addition to the above, all flat scanners exhibit large area crosstalk which is caused by reflected light from areas of a scanned image adjacent to the point being scanned. This can cause the X,Y, and or Z portions of CIE light to vary as much as 20%. You can get over 6 dE differences in L* alone under some circumstances and this cannot be corrected by profiling.

[Rhossydd]

Ditto and I've tried a number of such products over the years. Always awful. Because a scanner is absolutely the wrong tool for this task. A Spectrophotometer is.
  Wrong tool for the job!

Same experience here.
Scanners are great for scanning images, but no substitute for a spectrophotometer.

Don't waste any more time on it, just either buy a spectro or get someone to make you a custom profile.

[GWGill]

Using a scanner to create a printer profile is an exercise in futility.

Agreed - it's an interesting exercise, but typically doesn't give a good result.

The circumstances where it can give an acceptable result are pointless - you need to
create a scanner test chart using the printer in question, and to do this you need to
measure the scan chart reference values with a color measurement instrument.

(I'll edit the ArgyllCMS tutorial information to make this point more strongly.)

[Doug Gray]

Agreed - it's an interesting exercise, but typically doesn't give a good result.

The circumstances where it can give an acceptable result are pointless - you need to
create a scanner test chart using the printer in question, and to do this you need to
measure the scan chart reference values with a color measurement instrument.

(I'll edit the ArgyllCMS tutorial information to make this point more strongly.)

Yep, and if you can make a good print of a target, and measure it with a spectro, you can make a moderately decent scanner profile that works reasonably well for scanning prints from the same printer or ones with similar ink spectral characteristics. It still has dE76's of a bit over 1 average on independent, randomized, patches. Even with thousands of target patches.  However, with the program I did for the V850 scanner which removes most of the large area crosstalk, the same profile target produced profiles with an average dE under .5 compared to the spectro. As a further test, I made a target on US letter glossy with 2.3k patches, scanned it, fixed up the crosstalk, and made a printer profile from it. Worked quite well.

The large area crosstalk fixup has to model the amount of light re-reflected from nearby area and that varies with scanners. It's a somewhat involved process to model but once done it can be applied to any scanner of the same model or construction.

Of course none of that helps the scanner accurately scan anything other than prints. Testing it on my 9800 and 9500 against spectro measured results showed only slightly higher dE which suggests the CYM inks had fairly similar spectral distributions.

[digitaldog]

Maybe he should try using his Canon 100D (that he found out just today, actually produces a raw), or his old iPhone. 
https://forum.luminous-landscape.com/index.php?topic=131823.msg1126064#msg1126064
That landed like a lead balloon.

[Doug Gray]

Maybe he should try using his Canon 100D (that he found out just today, actually produces a raw), or his old iPhone. 
https://forum.luminous-landscape.com/index.php?topic=131823.msg1126064#msg1126064
That landed like a lead balloon.

Yeah, I was a bit surprised that he was unaware of that. RAW files are essential to get the requisite linear sensor info unmolested. Especially as he stated having a background with image sensors.

[digitaldog]

Especially as he stated having a background with image sensors.

If you believe that, I'm happy to sell you Greenland. 😈

[Doug Gray]

If you believe that, I'm happy to sell you Greenland. 😈

Yeah. no kidding. The dissonance between claiming sensor knowledge and not knowing his D100 had RAW capability was stark. It didn't seem to matter to him to learn this as he hasn't commented on it. Yet it was important enough to mention in his post.  Just weird.

[Hale_JP]

You know, speaking with you guys is like speaking with golden Billiken. You are shiny and arrogant, smirking in the face. But there's an iron bucket noise instead of words, and an air in the hollow head.

1)I know about sensors more than most of you.
But colorimetry has nothing to do with sensors. It is about very specific standards and methods of data conversion.
2)Making a profile (or calibrating, that depends on the process) does not necessary mean making a perfect profile. It means making it better than WITHOUT.
3)Scanner is a weak instrument, no question. But it is better than nothing. And it has excess specs comparing with mainstream LCD.
4) I am not preparing a data for publishing. I have a final viewing/home printing intents - it does not require absolute values and extra wide gammut.
The goal is maintaining visibly correct main tints - skin, air, water. Simple final consumer goals.
Colorimeters are another part of process.
5)Colorimeters are overpriced s**t. Spectrometers are cool. :-p
6)Yes I have 100D, first issue premium model. But it is still a toy camera. Who would ever use RAW on toy camera? I even forgot the feature. Well ML enables raw video... but who needs a video on a cam with mono mic, no smooth zoom and no sensor anti-shake.

Every thing has its intents, do not try mix intents!
Only idiots and indecently rich people spend K$ on toys and whistles for home appliances.

Printer: $25, scanner $15, ink $3/pack, paper $7/pack, IT8 target $20, monitor-laptop, or home-grade H-IPS
... colorimeter $500-$3000, WTF.

[digitaldog]

1)I know about sensors more than most of you.

Yet you didn't know the camera you claim to own, produced a raw file. Till Doug actually informed you of that sensor fact. Sad. 
“Nothing in all the world is more dangerous than sincere ignorance and conscientious stupidity” - Martin Luther King
The question YOU asked US:
Profiling printer, with scanner and Wolf's IT8. what am I doing wrong?

You don't listen or accept the answers provided by multiple knowledgeable people with experience to your questions. That's what's wrong to begin with!