Perceptual Rendering Argyll CMS

[digitaldog]

It's made up in the sense that I use it to refer to perceptual intent rendering that is reversible or has deterministic colorimetric rendering.

Thanks, it's what I suspected.

[Doug Gray]

Your bias and prejudice that Perceptual rendering isn't appropriate for you has been made clear more than once. The idea that there is no merit in any perceptual renderings from all products for all users is to be kind, hogwash. Again, profiles know nothing about color in context. I asked you in another thread why photographers shouldn't be given a choice in color rendering in E6 file (Velvia vs. Ektachrome as an example) which is no different than a Perceptual mapping of colors, you ignored that question. IF you need a colorimetric reproduction, Perceptual isn't for you. Yet many image makers desire pleasing color and that's exactly what a Perceptual rendering attempts, repeat, attempts to provide and not all are created equally. If you're quest is to argue with the author of ArgyllCMS he shouldn't be providing a Perceptual rendering in his product, I'm going to sit back and really enjoy that debate. 

I'm not biased against Perceptual and use it all the time for printing most of the pictures I take since the printer profile maker's idea of what looks good is usually acceptable. And I don't usually care that much whether they can be reproduced exactly in the future. If I do, it's easy enough to achieve that through reverse rendering using Relative Colorimetric into Adobe RGB and saving the file for that future case.

What I don't like about Perceptual is that it is ill defined, varies between vendors and settings, and requires that extra step in the last paragraph if I want to achieve future reproducibility.

If I'm putting much work into a photo, I'd rather tweak the tone curves and colors in Photoshop and use RC.

[digitaldog]

My current workflow uses Adobe Bridge, ACR and Photoshop CS6. However I will shortly be moving to Lightroom which I understand uses a variation on Prophoto colour space but with a gamma of 1.0. Is that colorspace supplied with the software as an icc/icm file that I can use as the source for building custom printer profiles to print direct from light room?

That's simply the underlying processing color space. You don't have to really address it at all. In LR, Melissa RGB is provided for the Histogram and numbers, that's ProPhoto RGB primaries with a 2.2 TRC (it's actually got a slight 'bump' down in the shadows like sRGB). WHY they did this is beyond me. If the time comes you want to know the actual numbers and histogram, setup a soft proof (type S key), that's what you'll get when you render the raw into that color space.

[digitaldog]

What I don't like about Perceptual is that it is ill defined, varies between vendors and settings, and requires that extra step in the last paragraph if I want to achieve future reproducibility.

Well the same could be said of RelCol. I provided an example of two profiles for the same device that produced hugely different output with the RelCol table. Everyone's color engine is different. Not all profiles, tables and results are equal.

[DaveRichardson]

When one supplies the requested picture, does the mapping software examine the gamut of colors within the image or the gamut of the color space (container for the image)? If the latter, one would not really need a picture, but merely a reference to the color space.

Hi Bill - Argyll asks for a source colour space -rather than a picture - when building a profile for output.

Sorry, I wasn't totally clear. The settings I referred to were from i1Profiler.

No worries Andrew - I did get that. i1match was even more restricted.

In LR, Melissa RGB is provided for the Histogram and numbers, that's ProPhoto RGB primaries with a 2.2 TRC (it's actually got a slight 'bump' down in the shadows like sRGB).

I'll supply that to the Argyll CMS when building the profiles to print.

Thanks all - I'm learning a lot here

Dave

Dave

[digitaldog]

I'll supply that to the Argyll CMS when building the profiles to print.

You can easily build a variant of ProPhoto RGB with a 1.0 simplified gamma in Photoshop's Color Settings itself. But I'm not sure you need to, best get feedback from Graeme. If all his software needs is the gamut of what you'll eventually render (into ProPhoto RGB), then ProPhoto RGB should work. As to the 'colors' that don't exist in ProPhoto RGB, not sure how this will all pan out.

[Doug Gray]

Well the same could be said of RelCol. I provided an example of two profiles for the same device that produced hugely different output with the RelCol table. Everyone's color engine is different. Not all profiles, tables and results are equal.

Yes, I've seen broken profiles too. So what.  Any compliant profile renders in-gamut colors via RelCol quite well. Out of gamut colors have to be mapped somewhere to the gamut boundary that is closest in some sense but the way vendors do that varies.

The same cannot be said of Perceptual which changes tone and colors and, since most images have most of their colors in-gamut this can be an issue.

[digitaldog]

Yes, I've seen broken profiles too.

Broken? How would you prove that? Again, not all profiles or rendering intents are created equally. And the Perceptual rendering is fair game to use after soft proofing the image in context and deciding as the image creators, you prefer that rendering. Velvia isn't better than Ektachrome, it's different and only better IF you subjectively prefer it.

[Doug Gray]

Broken? How would you prove that? Again, not all profiles or rendering intents are created equally. And the Perceptual rendering is fair game to use after soft proofing the image in context and deciding as the image creators, you prefer that rendering. Velvia isn't better than Ektachrome, it's different and only better IF you subjectively prefer it.

Give me a break.  RI is well defined in gamut. A profile that renders an L=1 on medium that has a minimum L=20, and, when reversed, claims to have rendered L=1, is broken.  Badly broken.  AtoB1 tables are supposed to accurately colorimetrically represent the printed color. Always. They are not supposed to scaled to the black point or in any way other than the media's white point.

[DaveRichardson]

You can easily build a variant of ProPhoto RGB with a 1.0 simplified gamma in Photoshop's Color Settings itself

I'd forgotten you could do that  . Must be my age - now re-learning as well as leaning !

Dave

[digitaldog]

Give me a break.  RI is well defined in gamut. A profile that renders an L=1 on medium that has a minimum L=20, and, when reversed, claims to have rendered L=1, is broken.  Badly broken.  AtoB1 tables are supposed to accurately colorimetrically represent the printed color. Always. They are not supposed to scaled to the black point or in any way other than the media's white point.

IF you believe that all profiles using a RelCol intent produce the same or even similar results, you need to test more than one profile package. Here's the same measured spectral data sent to two packages (Copra and i1Profiler) and the differences in the two (using Apply Image>Subtract) from a RelCol conversion. Yes I can get more specific with dE values but the point is, not all RelCol is equal!
Which is broken?

[Doug Gray]

IF you believe that all profiles using a RelCol intent produce the same or even similar results, you need to test more than one profile package.

What part of what I wrote do you not understand? I have pointed out that correct profiles can and do render out-of-gamut colors differently. They map those colors to gamut boundaries based on whatever they think appropriate. It's not specified by the ICC. I have also pointed out what I have determined to be a bad profile. Not just bad looking but a profile that wasn't made correctly which is defined only in AtoB1 and BtoA1, aka RelCol. My standard profiles are currently made with PM5 and I1Profiler. Both do a fine job with in-gamut colors using RelCol.

Yes I can get more specific with dE values but the point is, not all RelCol is equal!

Please do. As I said, I've seen bad, not just ugly, profiles that were not made based on ICC requirements. Of course that was looking at RelCol which is well defined by ICC for in gamut colors.

[Doug Gray]

Here's the same measured spectral data sent to two packages (Copra and i1Profiler) and the differences in the two (using Apply Image>Subtract) from a RelCol conversion.

Please supply the two profiles and I'll tell you which is bad. It's not rocket science.

[digitaldog]

What part of what I wrote do you not understand?

I guess this part: AtoB1 tables are supposed to accurately colorimetrically represent the printed color. Always.

My standard profiles are currently made with PM5 and I1Profiler. Both do a fine job with in-gamut colors using RelCol.

But accurately, colorimetrically, always? I suspect not.

[digitaldog]

Here's a MacBeth in sRGB, the difference in RelCol between the two profiles (again, not the same).

[digitaldog]

Here's a MacBeth in sRGB, the difference in RelCol between the two profiles (again, not the same).

And a dE report (notice max dE?)



--------------------------------------------------


dE Report


Number of Samples: 169000


Delta-E Formula dE2000


Overall - (169000 colors)
--------------------------------------------------
Average dE:   1.07
    Max dE:   3.01
    Min dE:   0.04
 StdDev dE:   0.63


Best 90% - (152099 colors)
--------------------------------------------------
Average dE:   0.95
    Max dE:   1.93
    Min dE:   0.04
 StdDev dE:   0.54


Worst 10% - (16901 colors)
--------------------------------------------------
Average dE:   2.15
    Max dE:   3.01
    Min dE:   1.93
 StdDev dE:   0.17


--------------------------------------------------

[Doug Gray]

I guess this part: AtoB1 tables are supposed to accurately colorimetrically represent the printed color. Always.But accurately, colorimetrically, always? I suspect not.

You are wrong. Accurate, reproducible color is extremely important to me. I check RelCol accuracy using a spectrophotometer. I make a random patch set using selections different from the stepped RGB values used for creating profiles. My prints that are in gamut match within the ability of people being able to tell that two, side by side, prints were made with different printers*. Or PM5 v I1Profiler for that matter.

*With the exception of artifacts like bronzing from specular reflection. Illuminated only at an angle that doesn't reflect back to viewer, they match. And they should match.

[digitaldog]

You are wrong. Accurate, reproducible color is extremely important to me. I check RelCol accuracy using a spectrophotometer

I suggest you check with your actual ICC profiles as I just did. 
Sorry if the reality of differences in RelCol in differing packages (from the identical spectral data) is ruining your life.

[digitaldog]

I check RelCol accuracy using a spectrophotometer.

What Spectrophotometer would that be, since you brought that into the discussion?

[Doug Gray]

And a dE report (notice max dE?)



--------------------------------------------------


dE Report


Number of Samples: 169000


Delta-E Formula dE2000


Overall - (169000 colors)
--------------------------------------------------
Average dE:   1.07
    Max dE:   3.01
    Min dE:   0.04
 StdDev dE:   0.63


Best 90% - (152099 colors)
--------------------------------------------------
Average dE:   0.95
    Max dE:   1.93
    Min dE:   0.04
 StdDev dE:   0.54


Worst 10% - (16901 colors)
--------------------------------------------------
Average dE:   2.15
    Max dE:   3.01
    Min dE:   1.93
 StdDev dE:   0.17


--------------------------------------------------

Obviously not the result of a spectro measurement. That would be useful to see which is most accurate.

Looks like a report of the entire gamut accuracy consistency using BtoA1 andAtoB1 processing of a print image. What print image? Was that the ColorChecker image?  If so, why not just measure the printed patches with a spectro to compare the profile accuracies? The larger dE2k values cluster around the gamut boundaries. Can't tell anything other than the profiles are slightly different. Unless the White Point was shifted, which can be noticeable, I don't believe any RelCol, in gamut print with those numbers, would be very easy to discern at all.