Pages: [1] 2 3   Go Down

Author Topic: Ideal V.S. Reality of linear behaviour  (Read 10608 times)

aaronchan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 617
Ideal V.S. Reality of linear behaviour
« on: October 21, 2017, 05:58:37 am »

Dear all,

I'm currently working with the latest Epson SC-P7000 and trying to see how linear the printer could after using my own custom profile.
My profiling target has 3075 patches, it is based on a 2744 generated by i1Profiler and added extra neutral and near neutral color into it.
The paper that I tested on was Hahnemuhle Photo Rag Pearl with PK ink

After the calibration and profiling, I've printed a 21 steps target with my profile and measured the test chart with i1Pro2
And here is the result as you can see on the attached image.
What surprised me is the printer does not act as linear as I thought compare to it's ideal linear target.
I have printed out an test image which I had been using for quite some time already. The test image looks perfectly fine as it's supposed be.
But technically, is it possible to make the printer more "linear"?

Thanks
Aaron

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Re: Ideal V.S. Reality of linear behaviour
« Reply #1 on: October 21, 2017, 10:29:24 am »

With the commercial or designer edition, the optional spectro-proofer and the Fiery software, you can re-linearize the printer. I'm not sure whether this can also be done without any one of these components, but a call to Epson would clarify that. That said, with the SC-P5000 I'm using, it may not be necessary. I've produced custom profiles and conducted many of the kinds of tests you show; results vary from being highly satisfactory to less satisfactory depending on the paper and the profiling approach. Using high quality gloss/luster papers, the correct media type, best M condition and suitable patch set (for example the 1877 set), I've gotten this same kind of "results" line to track the "ideal" line very closely. You'll find that in some of the reviews I've published here. And this is without re-linearizing the printer.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Alan Goldhammer

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4344
    • A Goldhammer Photography
Re: Ideal V.S. Reality of linear behaviour
« Reply #2 on: October 21, 2017, 12:09:07 pm »

I don't use iProfiler but ArgyllCMS permits me to add any number of B/W patches to the target set.  I routinely use a two step process, a preconditioning set of patches and once that first profile is done, I refine it with a larger patch set consisting of near neighbor patches AND a 51 step B/W patch set.  Because I have the B/W patch set included, this allows the linearization to take place during the profiling process.  My question is whether iProfiler allows one to do this.  It should lead to a linear B/W print out following profiling, correct?
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Ideal V.S. Reality of linear behaviour
« Reply #3 on: October 21, 2017, 12:26:17 pm »

Dear all,

I'm currently working with the latest Epson SC-P7000 and trying to see how linear the printer could after using my own custom profile.
My profiling target has 3075 patches, it is based on a 2744 generated by i1Profiler and added extra neutral and near neutral color into it.
The paper that I tested on was Hahnemuhle Photo Rag Pearl with PK ink

After the calibration and profiling, I've printed a 21 steps target with my profile and measured the test chart with i1Pro2
And here is the result as you can see on the attached image.
What surprised me is the printer does not act as linear as I thought compare to it's ideal linear target.
I have printed out an test image which I had been using for quite some time already. The test image looks perfectly fine as it's supposed be.
But technically, is it possible to make the printer more "linear"?

Thanks
Aaron

No idea what the "21 steps target" is from. I assume some sort of supplied tif or jpeg image. Printing targets should not be done with BPC. You shouldn't normally even use Rel Col. If you want patch colors to match use Abs Col.

I created a 21 wedge chart in 16 bit tiff with the values from your spreadsheet and included the tif file as well as a plot  showing the original L* and printed L*.  I used Costco paper on a Canon 9500 which is limited on both ends. The lowest printable is 7, the highest 95. You can see where the two ends diverge from the ideal in the plot.

Try printing the tif file using AbsCol and measure the patches. I'd bet it would be quite close.
« Last Edit: October 21, 2017, 12:30:53 pm by Doug Gray »
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Re: Ideal V.S. Reality of linear behaviour
« Reply #4 on: October 21, 2017, 01:21:06 pm »

Aaron, I'm assuming you used ABSCOL Rendering Intent, otherwise I would have expected there to be more separation between the lines at the low end of the tone scale. But Doug is correct, for proofing you want to perform this test in Absolute Rendering Intent. That automatically disables Black Point Compensation as well.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

aaronchan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 617
Re: Ideal V.S. Reality of linear behaviour
« Reply #5 on: October 21, 2017, 02:45:29 pm »

No idea what the "21 steps target" is from. I assume some sort of supplied tif or jpeg image. Printing targets should not be done with BPC. You shouldn't normally even use Rel Col. If you want patch colors to match use Abs Col.

I created a 21 wedge chart in 16 bit tiff with the values from your spreadsheet and included the tif file as well as a plot  showing the original L* and printed L*.  I used Costco paper on a Canon 9500 which is limited on both ends. The lowest printable is 7, the highest 95. You can see where the two ends diverge from the ideal in the plot.

Try printing the tif file using AbsCol and measure the patches. I'd bet it would be quite close.

Hi Doug and Mark,

On my test, I have printed with Rel. Col with BPC on, which may not be correct.
I will go back to try with the Abs Col and see what will happen.

The reason why I want to point this out is not because I want to do proofing.
But I just want to understand if an Epson driver with RGB driver and proper custom icc profile,
will it be able to produce absolute linear B&W print.

Thanks
Aaron

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Re: Ideal V.S. Reality of linear behaviour
« Reply #6 on: October 21, 2017, 05:36:46 pm »

Hi Doug and Mark,

On my test, I have printed with Rel. Col with BPC on, which may not be correct.
I will go back to try with the Abs Col and see what will happen.

The reason why I want to point this out is not because I want to do proofing.
But I just want to understand if an Epson driver with RGB driver and proper custom icc profile,
will it be able to produce absolute linear B&W print.

Thanks
Aaron

OK, here it becomes necessary to sort out purposes and methods. If you are trying to see whether a printer can accurately reproduce the file values using the profile you are using, this is proofing and you do need to use ABSCOL so that no application-induced transformations are taking place. ABSCOL is designed to reproduce the values in the image file without making adjustments. RELCOL with BOC remaps pixel values so this is not a correct test of profile and printer behavioural accuracy. Step One is to determine whether the printer can accurately reproduce those file values. If it can, Step Two is to see what happens to them once you introduce another Rendering Intent with BPC. Depending on the paper, the differences could be either very little or very considerable. Please see my articles on this for further elaboration. If Step One shows the printer cannot reproduce the file values quite accurately, then the problem is up-stream of the Rendering Intent/BPC business and needs to be addressed.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Ideal V.S. Reality of linear behaviour
« Reply #7 on: October 21, 2017, 06:24:27 pm »

OK, here it becomes necessary to sort out purposes and methods. If you are trying to see whether a printer can accurately reproduce the file values using the profile you are using, this is proofing and you do need to use ABSCOL so that no application-induced transformations are taking place. ABSCOL is designed to reproduce the values in the image file without making adjustments. RELCOL with BOC remaps pixel values so this is not a correct test of profile and printer behavioural accuracy. Step One is to determine whether the printer can accurately reproduce those file values. If it can, Step Two is to see what happens to them once you introduce another Rendering Intent with BPC. Depending on the paper, the differences could be either very little or very considerable. Please see my articles on this for further elaboration. If Step One shows the printer cannot reproduce the file values quite accurately, then the problem is up-stream of the Rendering Intent/BPC business and needs to be addressed.

Mark,
I suspect the problem is upstream and that the 21 step neutral wedges do not have the L* shown in his spreadsheet.  21 step "linear" in L* wedges would normally go from L0 to L100 in steps of 5. Printing them using RelCol, but not BPC, should produce values that clip at the black point but otherwise match pretty closely all the way up to 100. RelCol values are scaled to the white point but clipped at the blackpoint by ICC compliant profiles and I1Profiler does this.

What I think happened is he printed some 21 step set that is scaled against something other than L*. Then, he measured the dark patch and light patch and drew a line between them. I'd say that was 99.9% likely since the starting and ending values match to the last fractional decimal place. It could have been something like a sRGB tone curve or something else with a gamma around 2.2. The L* isn't exactly a gamma curve but it's approx. 2.5 and his plot indicates the 21 step image used a lower one.

This is why I attached a tif file that has exactly those same starting and ending points and equal L* steps in between. This has the advantage of being within his printer's gamut. If he downloads and prints that using AbsCol then measures them he should have results very close to a straight line if his profile is good.
« Last Edit: October 21, 2017, 06:28:40 pm by Doug Gray »
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Re: Ideal V.S. Reality of linear behaviour
« Reply #8 on: October 21, 2017, 06:37:26 pm »

Plausible Doug - maybe I'm assuming more about his methodology than I should (at first I assumed he was in ABSCOL, he wasn't; now I'm assuming he's working in Lab and using the L* scale - perhaps he isn't). He should give us the details.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Re: Ideal V.S. Reality of linear behaviour
« Reply #9 on: October 21, 2017, 06:38:16 pm »

Oh - and I forgot to mention - it is also good to not only focus on L*, but also look for any meandering from 0 values of a* and b*.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Jeff-Grant

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 568
    • http://www.jeff-grant.com
Re: Ideal V.S. Reality of linear behaviour
« Reply #10 on: October 21, 2017, 10:21:45 pm »


The reason why I want to point this out is not because I want to do proofing.
But I just want to understand if an Epson driver with RGB driver and proper custom icc profile,
will it be able to produce absolute linear B&W print.


If you want linear B&W, you would be a lot better off using QTR and its linearising tools.
Logged
Cheers,
 Jeff  www.jeff-grant.com

Alan Goldhammer

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4344
    • A Goldhammer Photography
Re: Ideal V.S. Reality of linear behaviour
« Reply #11 on: October 22, 2017, 09:33:33 am »

If you want linear B&W, you would be a lot better off using QTR and its linearising tools.
Yes, that is of course true but I think the OP wants to know if through a normal CM workflow he can achieve a linear response.  I know I can using the profiling method I described earlier in this thread.  QTR does require additional steps and of course one can use the Epson ABW print driver as well (with softproofing under WinOS if needed).
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 20677
  • Andrew Rodney
    • http://www.digitaldog.net/
Re: Ideal V.S. Reality of linear behaviour
« Reply #12 on: October 22, 2017, 10:39:57 am »

If you want linear B&W, you would be a lot better off using QTR and its linearising tools.
Well certainly NOT an Epson driver and hopefully a product that does allow linearization. Of course asking WHY we need this and to what degree is an important question too!
Logged
http://www.digitaldog.net/
Author "Color Management for Photographers".

aaronchan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 617
Re: Ideal V.S. Reality of linear behaviour
« Reply #13 on: October 22, 2017, 01:04:12 pm »

Sorry guys!

Thanks for all of your inputs.
Just had a busy Sunday in China.
I'll get back to the office and try to explain my workflow a bit more
Plus I'll try to fix my testing method as well since there might be some problem on my previous test

I'll update you guys soon

Best Regard,
Aaron

Alan Goldhammer

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4344
    • A Goldhammer Photography
Re: Ideal V.S. Reality of linear behaviour
« Reply #14 on: October 22, 2017, 01:34:35 pm »

Well certainly NOT an Epson driver and hopefully a product that does allow linearization. Of course asking WHY we need this and to what degree is an important question too!
QTR offers two approaches.  One can use the QTR driver to print B/W and of course this bypasses the Epson driver.  The second approach is the one I alluded to and that is to use the Epson ABW print driver and prepare a profile that linearizes the print output using the QTR script tools.  As I noted this will only work with a WinOS system as MacOS disabled this feature five years ago.  The other approach is the one I use by including a B/W patch set when I create normal ICC profiles with ArgyllCMS
Logged

aaronchan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 617
Re: Ideal V.S. Reality of linear behaviour
« Reply #15 on: October 23, 2017, 11:55:55 am »

Hi Guys,

Finally today I can go back to my office and do some test.
Thanks Doug for providing me the idea.

So here is my test result.
i've used what Doug gave me and printed with Absolute Col rendering intent.

The result is very linear compare to the ideal line, which formed based from the blackest point and the paper white point.

So after this experiment, I have 2 more questions regarding to this:

1st question:
Doug provided me this special target and I understand how does he created this. Basically 21 steps are plot from my ideal numbers from excel.
But when map my black point and white point to a "normal 21 steps" target (0-255), how come I could not get the middle patches RGB number as same as my excel calculation?
For example, on excel sheets, the middle patch RGB number should be 121,121,121, in Photoshop, it will become 131,131,131?

2nd question:
The reason I'm doing this test is because I'm trying do digital negatives.
In theory, a more linear we can get, a better control we could have during the process.
You may not agree but at least this is so far working good with our workflow.
But I doubt it I should use absolute col. rendering intent when I print my digital neg.
What we've been using is Rel Col with BPC on.
The image looks nice but when we print the 21steps target, it turns out not very linear.
So, any ideas?

Of course I'll do some more test on my side, would like to share more and update you guys.

Thanks everyone!

Regard,
Aaron

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Ideal V.S. Reality of linear behaviour
« Reply #16 on: October 23, 2017, 05:08:19 pm »

Hi Guys,

1st question:
Doug provided me this special target and I understand how does he created this. Basically 21 steps are plot from my ideal numbers from excel.
But when map my black point and white point to a "normal 21 steps" target (0-255), how come I could not get the middle patches RGB number as same as my excel calculation?
For example, on excel sheets, the middle patch RGB number should be 121,121,121, in Photoshop, it will become 131,131,131?
You can't expect Lab values, in particular L*, the luminance component of Lab, to match other colorspace values. They don't. For instance Adobe RGB, ProPhoto RGB, and sRGB all will have different numbers for exactly the same gray. Further, there are different 21 step gray wedges. Classic ones include one for film transmission where the lightest patch transmits the clear base, the next lightest transmits 70.7% of that, and each subsequent one transmits 70.7% of prior patch. This provides a thousand to one dynamic range which corresponds to a DMax of 3,  a larger DMax than virtually any printer, even specialized ones, though film can do it.  Other sets are based on B&W with some dot gain, or an RGB space like sRGB.  Lab is a sort of universally used measure because it is the common color space used by all ICC printer profiles.

Quote
2nd question:
The reason I'm doing this test is because I'm trying do digital negatives.
In theory, a more linear we can get, a better control we could have during the process.
You may not agree but at least this is so far working good with our workflow.
But I doubt it I should use absolute col. rendering intent when I print my digital neg.
What we've been using is Rel Col with BPC on.
The image looks nice but when we print the 21steps target, it turns out not very linear.
So, any ideas?
Also, wedge patches aren't linear other than perhaps some used in laboratories. Linear ones look awful with the lighter side looking way to bright. This is because the eye's perception is non-linear.

As an example, L*=10 is 5 times more luminous (in a scientific sense of the reflected percentage of light) than L*=2.  And that indeed is linear.  But it changes. L*=50 is about 18, not 5, times more luminous than L*=10. This is because L* transitions to a heavily non-linear response in the early teens but is linear at lower levels.

Other colorspaces and B&W dot gain responses differ from this.

The word "linear" and "linearize" are bandied about but typically refer to matching some inking response like in prepress. Beware using it more broadly. In most cases it doesn't mean what it implies.

You can load up the 21 wedge image in Photoshop and see what the Lab values of each patch are with the info tool. You can view->proof other color spaces and show their RGB values at the same time. This can be interesting and give you an idea why these seem to vary.
« Last Edit: October 23, 2017, 05:14:44 pm by Doug Gray »
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Re: Ideal V.S. Reality of linear behaviour
« Reply #17 on: October 23, 2017, 05:16:06 pm »

Aaron - your diagram now looks like what I would expect for that paper, so the switch to ABSCOL was important to achieving that result and demonstrates that your printer/profile response to the file values preserves a high degree of linearity - defined here as input matching output in a grayscale with L* intervals of 5 from Black to White or vice versa. Once you change the Rendering Intent to RELCOL with BPC, there will be some remapping, cause some departure from the closeness of the match you now obtained particularly at the Black end of the scale, but the path up the scale should remain quite linear, albeit at a different angle. I've described and illustrated all this in several of my articles on this website.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2197
Re: Ideal V.S. Reality of linear behaviour
« Reply #18 on: October 23, 2017, 05:30:17 pm »

Aaron - your diagram now looks like what I would expect for that paper, so the switch to ABSCOL was important to achieving that result and demonstrates that your printer/profile response to the file values preserves a high degree of linearity - defined here as input matching output in a grayscale with L* intervals of 5 from Black to White or vice versa. Once you change the Rendering Intent to RELCOL with BPC, there will be some remapping, cause some departure from the closeness of the match you now obtained particularly at the Black end of the scale, but the path up the scale should remain quite linear, albeit at a different angle. I've described and illustrated all this in several of my articles on this website.

Good advice Mark. I tend to get too mired in techie detail. Looks to me like he's good to go with that profile and RelCol with BPC likely will provide him good results. Though I'm just a bit cautious because of his reference to making some sort of "negative."  I'm not sure what he's referring to and don't know whether this would meet his needs.
Logged

Jeff-Grant

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 568
    • http://www.jeff-grant.com
Re: Ideal V.S. Reality of linear behaviour
« Reply #19 on: October 23, 2017, 06:08:46 pm »

Not wishing to sound like a broken record but QTR rather the Epson driver is the more common way to produce digital negatives AFAIK.
Logged
Cheers,
 Jeff  www.jeff-grant.com
Pages: [1] 2 3   Go Up