Pages: [1]   Go Down

Author Topic: Different Relative Colorimetric Rendering?  (Read 1048 times)

Brad Paulson

  • Full Member
  • ***
  • Offline Offline
  • Posts: 242
Different Relative Colorimetric Rendering?
« on: January 29, 2018, 05:34:01 PM »

I don't know how long it will last, but lately I've become more a fan of Relative Colorimetric over Perceptual. 

The challenge of course is managing clipped colors.

I've read a few web pages that state while the ICC standards proscribe a hard clip where an image gamut exceeds a print gamut, some manufactures fudge the line and use some proprietary perceptual-like algorithm to differentiate colors that would be clipped.  One web page, for example, stated that Argyll (at least at some point in time) observed a hard clip consistent with the wishes of the ICC, while Photoshop and a few other software manufactures deployed some methodology to differentiate clipped colors.  I can dredge that page up later if someone wants to see it.

As a practical matter, there are many ways to manually massage image data into print gamut, but at least for the last few iterations of doing that it seems like what is described above could be an attractive idea.  That assumes that these mutterings I'm reading are true . . . 

So a couple questions come to mind for which I can't easily find answers.

1. Is it true that some implementations of RelCol do massage clipped data?

2. Which software platforms do this?

3. How (presumably hue, saturation and/or luminosity, that would be nice to know)?
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 11519
    • http://www.markdsegal.com
Re: Different Relative Colorimetric Rendering?
« Reply #1 on: January 29, 2018, 05:39:55 PM »

I think the best way of dealing with clipped data is to first decide whether the data being clipped really matters to the quality of the photo. If it does, bring as much of it as sensible into gamut in your image editing software using luminance and saturation controls, watching both the histogram and the photo itself as you proceed. Once you have that the way you like it, I don't think the Rel Col recipe will matter much.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Brad Paulson

  • Full Member
  • ***
  • Offline Offline
  • Posts: 242
Re: Different Relative Colorimetric Rendering?
« Reply #2 on: January 29, 2018, 06:35:36 PM »

Thanks Mark.  That's the first workflow of course.  But at some point we reach diminishing returns.  What I'm really asking if there's another RelCol tool in the IQ arsenal.  I'm working with 4' long prints and at that scale every leaf in the forest is visible.  If one RC implementation can help over another, that's what I'm seeking to understand.
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1346
Re: Different Relative Colorimetric Rendering?
« Reply #3 on: January 29, 2018, 07:05:24 PM »

Thanks Mark.  That's the first workflow of course.  But at some point we reach diminishing returns.  What I'm really asking if there's another RelCol tool in the IQ arsenal.  I'm working with 4' long prints and at that scale every leaf in the forest is visible.  If one RC implementation can help over another, that's what I'm seeking to understand.

I agree with Mark. And using Rel. Col. W/O BPC makes soft proofing more clear in many ways. All in-gamut colors will be unchanged when you toggle on/off soft proof (ctrl y) if you have show blacks checked but not show paper color. This makes it easy to see if a color is exceeding gamut or not as you will see a shift.

There is one potential problem area though. Your monitor may not be able to show colors that are printable but out of the monitor's gamut. The best way to find out is to temporarily desaturate by 20% in the color settings (40% if you don't have a wide gamut monitor). This expands the gamut the monitor will display to encompass colors you can print. Of course the image is desaturated somewhat so it's not an overall realistic view. But doing this allows you, by alternate ctrl y keystrokes, to visually see if any colors are being shifted because they are outside the printer's gamut.

It's really a good idea to bring all colors in the image into Rel. Col. gamut because then you will have the most  reproducible image with other printers and similar papers. Otherwise you are at the mercy of whatever the profile creator decided to map the out of gamut colors to and it is completely unspecified.
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 11519
    • http://www.markdsegal.com
Re: Different Relative Colorimetric Rendering?
« Reply #4 on: January 29, 2018, 07:45:40 PM »

.................... Otherwise you are at the mercy of whatever the profile creator decided to map the out of gamut colors to and it is completely unspecified.

Doug, this is an interesting point - I was given to understand that one of the main advantages of RelCol from a technical perspective is that the mapping is defined by an ICC specification, (unlike Perceptual), and according to that specification OOG colours are to be mapped to their nearest value on the gamut boundary. Presumably that is a mathematically determinate positioning in every case, or are you suggesting that perhaps it is not and that the manner in which this is calculated can be handled differently by different profiling apps? 
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1346
Re: Different Relative Colorimetric Rendering?
« Reply #5 on: January 29, 2018, 07:56:35 PM »

Doug, this is an interesting point - I was given to understand that one of the main advantages of RelCol from a technical perspective is that the mapping is defined by an ICC specification, (unlike Perceptual), and according to that specification OOG colours are to be mapped to their nearest value on the gamut boundary. Presumably that is a mathematically determinate positioning in every case, or are you suggesting that perhaps it is not and that the manner in which this is calculated can be handled differently by different profiling apps?
Nope. Rel. Col. is specified only for in gamut colors. There are no requirements on mapping out of gamut colors. There is only advice that out of gamut colors should be unsurprising. For instance mapping an OOG cyan to red would be legal but probably surprising. However, there is no technical definition of unsurprising. I think most profile vendors map to the nearest dE76 on the gamut surface but don't know for certain.

Perceptual is quite a different beast since even the concept of out of gamut doesn't apply. Or in gamut for that matter.
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 11519
    • http://www.markdsegal.com
Re: Different Relative Colorimetric Rendering?
« Reply #6 on: January 29, 2018, 08:16:35 PM »

Nope. Rel. Col. is specified only for in gamut colors. There are no requirements on mapping out of gamut colors. There is only advice that out of gamut colors should be unsurprising. For instance mapping an OOG cyan to red would be legal but probably surprising. However, there is no technical definition of unsurprising. I think most profile vendors map to the nearest dE76 on the gamut surface but don't know for certain.

Perceptual is quite a different beast since even the concept of out of gamut doesn't apply. Or in gamut for that matter.

Yes, I think that's correct based on the content of an ICC paper on V4 profiles.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Brad Paulson

  • Full Member
  • ***
  • Offline Offline
  • Posts: 242
Re: Different Relative Colorimetric Rendering?
« Reply #7 on: January 30, 2018, 12:00:45 PM »

Nope. Rel. Col. is specified only for in gamut colors. There are no requirements on mapping out of gamut colors. There is only advice that out of gamut colors should be unsurprising. For instance mapping an OOG cyan to red would be legal but probably surprising. However, there is no technical definition of unsurprising. I think most profile vendors map to the nearest dE76 on the gamut surface but don't know for certain.

Thanks for the desaturation tips Doug — forgot about that one.  When I finish processing a photo it’s almost always in Photoshop, and when I see OOG colors I can’t easily or happily resolve, my preferred method is to save a downsized “deleteme.tif” copy of the image file, load it up in ColorThink and look at the image datapoints that lay outside the printer gamut.  That gives me a great idea where the problem colors lie in terms of luminosity, saturation and hue and the spaces I have available to shift the problem areas around. I then typically use Tony Kuyper’s luminosity masks and the hue panel to nudge those areas into compliance.

The histogram is easy enough, but the problem with the OOG warnings in PS is they are not accurate enough. A troublesome image that inspired my post here is a 4’ high picture of a sunlit bamboo forest I’m working on. At some point in the back and forth fine tuning everything, there is a point where balancing the luminosity and saturation (hue doesn’t work well with this one) will degrade the overall IQ. I can see there is data in some of the leaves that I could tap into, but condensing the image’s luminosity or saturation comes at great pain and/or is practically impossible without doing damage on a larger scale. 

Anyway, I am going to reprocess this particular image today to preserve/amplify the luminosity differences accessible in my file (essentially by hand blending multiple versions of the processed RAW file, 16 bit 51MB) and that might be my only practical solution for this shot. I need to do that anyway.  The realization that I was headed that way yesterday made me wonder if different versions of RelCol might retain more detail in the leaves.  I can imagine, for example, some algorithm slightly amplifying luminosity differences between midtone OOG colors, something like a high pass filter, that would compensate in appearance at least in luminosity for more flattened chromaticity in the OOG color rendering. 

In any event, it seems like potentially a good idea at the moment to have something like that built into a RelCol rendering pipeline, and because I have a number of print pipelines residing on my computer it made me wonder about any known differences among manufacturers.  I know the answer to this question if there is one would yield almost imperceptible results, but in 4’ long prints, a few changes like this can be perceptible. 
« Last Edit: January 30, 2018, 12:22:50 PM by Brad Paulson »
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 14115
    • http://www.digitaldog.net/
Re: Different Relative Colorimetric Rendering?
« Reply #8 on: January 30, 2018, 01:18:31 PM »

Perceptual is like transparency film: Ektachrome, Agfachrome, Fujichrome; they all differ subjectively based upon what the manufacturers believe their customers will subjectively prefer.
ICC profiles and Histograms know nothing about color in context. Only individual color values. That isn't a limitation of humans. We soft proof and select the RI based on how all the millions of values co-exist to produce what appears to us to be an image. Within the limitations of the display gamut and other factors.
Logged
Andrew Rodney
Author “Color Management for Photographers"

Brad Paulson

  • Full Member
  • ***
  • Offline Offline
  • Posts: 242
Re: Different Relative Colorimetric Rendering?
« Reply #9 on: January 30, 2018, 01:50:48 PM »

Touché.  Nice point worthy of some thought. 
Logged
Pages: [1]   Go Up