Viewing Booth and Display

[BradFunkhouser]

I've gotten good results using ArgyllCMS camera profiles, printer profiles and absolute colorimetric (scene referred) conversions to get prints that match art originals, but getting my display to match the prints has always been more of a curiosity than a priority.  Now I'm working on it.  My photographic copy lighting setup can be configured as a viewing booth.  It's in front of and to the left of my display.  There are no other light sources in the room.  With the viewing booth lights on, the ambient light striking the display is 11 Lux.

Measuring the viewing booth lights at the print position plane using i1pro with ambient cap shows (averages over 5 readings using ArgyllCMS spotread -a):

   513 Lux
   Yxy  513.0   0.3463   0.3615
   CCT  4990K
   Closest Daylight temperature  4960K


My main proofing paper has no OBAs and has D50 Lab white point:    97.0   0.5   1.9    (Yxy  92.5  0.3495  0.3614)

I measured the display booth light reflecting from the proofing paper with ambient cap (from 10" away to avoid shadows, though the ambient cap seems like it will be getting some reading from the shadowed areas?):

   403 Lux
   Yxy  403.0   0.3525   0.3664
   CCT  4790K
   Closest Daylight temperature  4760K

Then I measured the display booth light reflecting from the proofing paper as emissive, with the ambient cap removed, using spotread -e  (from 10" away to avoid shadows):

   Yxy  153.0   0.3470   0.3600
   CCT  4960K
   Closest Daylight temperature  4950K


I have two other important papers, both non OBA, but their white points are slightly darker than my main proofing paper.  Based on this data, what cd/m^2  and white point values do you think I should target when adjusting the display, creating calibration curves, and profiling (and what math is going on to figure this out)?   Should I be measuring anything in some other way?    Thanks.

[BradFunkhouser]

Knowing the spectrum of the viewing light together with the luminance and spectrum of that light reflecting off the proofing paper,  it seemed like we should have some process to be able to compute targets for the display's luminance and white point for the standard observer.  But I see now that many things come strongly into play here... the border around the physical paper and around the white on the display, partial versus full adaptation, different types of adaptation (sensory versus cognitive) due to reflective versus self luminous surfaces, levels of illuminant discounting, etc.  etc.   

So we end up having to eyeball it.

I put a large dark neutral border (L25) around the proofing paper in the viewing booth, eliminated all other white objects from view so my eyes saw the proofing paper as the dominant white, adjusted a display border until it visually matched the dark border in the viewing booth, and then adjusted the display white box inside that border until I saw a match with the proofing paper.

This eyeballing yielded display calibration targets of 140 cd/m^2 with a daylight 5700k white point.

[Doug Gray]

Knowing the spectrum of the viewing light together with the luminance and spectrum of that light reflecting off the proofing paper,  it seemed like we should have some process to be able to compute targets for the display's luminance and white point for the standard observer.  But I see now that many things come strongly into play here... the border around the physical paper and around the white on the display, partial versus full adaptation, different types of adaptation (sensory versus cognitive) due to reflective versus self luminous surfaces, levels of illuminant discounting, etc.  etc.   

So we end up having to eyeball it.

I put a large dark neutral border (L25) around the proofing paper in the viewing booth, eliminated all other white objects from view so my eyes saw the proofing paper as the dominant white, adjusted a display border until it visually matched the dark border in the viewing booth, and then adjusted the display white box inside that border until I saw a match with the proofing paper.

This eyeballing yielded display calibration targets of 140 cd/m^2 with a daylight 5700k white point.

Good approach. Well thought out.

[BradFunkhouser]

To evaluate the differences between my viewing booth lights and D50 spectrum, I used spectral data for my printer target and spectral data for my lights to build a custom illuminant printer profile for my proofing paper.

For testing, I selected a set of scene referred art reproduction images with gamuts that fit inside those of both proofing paper and display, and I have existing D50 absolute prints for these images.  Before making prints using the custom illuminant profile, I wanted to use the two printer profiles to evaluate what the differences would be for these images when printed for one illuminant but viewed under the other.  How best to do that?

This is what I came up with...   Copy image and convert absolute to printer space with standard D50 profile (how the existing prints were produced).  Copy image and convert absolute to printer space with custom illuminant profile, then assign the D50 profile to that file so those printer device values are tagged to appear as they would under D50.  Bring those two files into ColorThink Pro worksheet, overlay a target grid to create a color list, compute Delta Es and sort to see which colors are most different and in what ways.

Is there anything wrong with this approach? 

What other ways are there to evaluate illuminant differences? 

Thanks.

[Doug Gray]

To evaluate the differences between my viewing booth lights and D50 spectrum, I used spectral data for my printer target and spectral data for my lights to build a custom illuminant printer profile for my proofing paper.

For testing, I selected a set of scene referred art reproduction images with gamuts that fit inside those of both proofing paper and display, and I have existing D50 absolute prints for these images.  Before making prints using the custom illuminant profile, I wanted to use the two printer profiles to evaluate what the differences would be for these images when printed for one illuminant but viewed under the other.  How best to do that?

This is what I came up with...   Copy image and convert absolute to printer space with standard D50 profile (how the existing prints were produced).  Copy image and convert absolute to printer space with custom illuminant profile, then assign the D50 profile to that file so those printer device values are tagged to appear as they would under D50.  Bring those two files into ColorThink Pro worksheet, overlay a target grid to create a color list, compute Delta Es and sort to see which colors are most different and in what ways.

Is there anything wrong with this approach? 

What other ways are there to evaluate illuminant differences? 

Thanks.

It's a good approach but CTP can be somewhat slow on a large image. CTP has a lot of useful features but speed doing dE comparisons isn't one of them.

Another approach is to convert each, then assign to the other profile. then you can compare those two images to the original to see how the prints made with one profile illuminant would look under the other.

I use Matlab with some scripts and functions I've written to look at these sorts of differences.

How much and where color shifts occur with different illuminants as adapted to D50 depends on the spectral characteristics of both the illuminant and inks. A useful metric for printer reviews would be color shift statistics relative to common illuminants including, especially, LED lighting as that is becoming quite popular. This would be useful for those who don't or can't make illuminate matched custom profiles and are just trying to get the best results they can with canned or outsourced custom profiles.

Needless to say, uV issues combined with paper OBs can produce even bigger differences.

Also, when you are using an illuminant that is not close to D50 the spectral characteristics of the artwork itself may cause color shifts. It's entirely possible that the shifts from the printer ink are less than those from the artwork.

You could have a print illuminated with the non D50 matching artwork illuminated with D50 side by side but they wouldn't match each other when viewed with the same illuminant, whether D50 or whatever.

Compensating for that requires spectral info on the artwork in addition to the illuminant.

[BradFunkhouser]

Needless to say, uV issues combined with paper OBs can produce even bigger differences.

Yes.  Definitely.  I only use non oba papers for matching work.

It's a good approach but CTP can be somewhat slow on a large image. CTP has a lot of useful features but speed doing dE comparisons isn't one of them.

Another approach is to convert each, then assign to the other profile. then you can compare those two images to the original to see how the prints made with one profile illuminant would look under the other.

I use Matlab with some scripts and functions I've written to look at these sorts of differences.

Converting to printer space then assigning the other profile does seem to work pretty well.  The average difference across the test images between my viewing booth lights and D50 was 1.3,  with the worst colors (light tans) maxing out just under 3.  I did go ahead and print a few of the test images using the custom illuminant profile for a side by side visual comparison with the D50 version in the viewing booth.  The DeltaE maps from ColorThink Pro for those images were good predictors of which areas would be noticeably different (though barely noticeable).  Yay, I like it when the model works!

I've done a lot of reading and experimentation with taking a scene referred art reproduction image file and matching a print from that file in a viewing booth with it's  image on a  display.  Here's a summary of my current understanding, and a question.  Please point out any problems you see...


Standard observer XYZ values respresent stimulus response only.  They tell us whether two different spectral stimuli will appear the same when compared side by side while isolated from all other visual stimuli.

CIE Lab extends that model with the definition of a white point and the use of chromatic adaptation functions to convert colors so they appear consistent, relative to each other, when the observer is adapted to different white points.  Lab is the beginning of an appearance model, but it assumes full adaptation to the white point and doesn't take into account appearance changes due to other factors such as background, surround, and luminance levels. 

The XYZ/Lab models work well for image capture and reproduction onto reflective media, but when we have a reflective viewing booth side by side with a self luminous display, we experience a breakdown in these models.  We can strive to get measured XYZ, white point, and Lab values that match between the two, but even when the measured values do match, we don't actually perceive a match.  This lack of a perceived match seems to go beyond what can be explained by instrument error or differences in individual observer.

In an effort to use the CIE XYZ/Lab models, we adjust the luminance of the viewing booth and/or the display to get a match while eyeballing changes to the image on the display until we perceive a white point match.  Then we build calibration curves for those target values and create a profile for converting an image into display space.  We can get a very close match between viewing booth and display using this technique, but that match is built on the eyeballing, so there's not really an accurate perceptual appearance model underpinning the effort.

CIE CAM02 takes into account levels of luminance, background, surround, and level of adaptation.  Given a paper white background in a viewing booth, a custom illumiinant printer profile for that paper under those lights, and a display that's been adjusted, calibrated and profiled for a white point that matches the measured Yxy values of the viewing booth lights reflecting off that paper white, can CAM02 be used to create a perceived match between print and display without any eyeballing adjustments required?

[Doug Gray]

Yes.  Definitely.  I only use non oba papers for matching work.

Converting to printer space then assigning the other profile does seem to work pretty well.  The average difference across the test images between my viewing booth lights and D50 was 1.3,  with the worst colors (light tans) maxing out just under 3.  I did go ahead and print a few of the test images using the custom illuminant profile for a side by side visual comparison with the D50 version in the viewing booth.  The DeltaE maps from ColorThink Pro for those images were good predictors of which areas would be noticeably different (though barely noticeable).  Yay, I like it when the model works!

Yes, that's consistent with what I see.

I've done a lot of reading and experimentation with taking a scene referred art reproduction image file and matching a print from that file in a viewing booth with it's  image on a  display.  Here's a summary of my current understanding, and a question.  Please point out any problems you see...


Standard observer XYZ values respresent stimulus response only.  They tell us whether two different spectral stimuli will appear the same when compared side by side while isolated from all other visual stimuli.

CIE Lab extends that model with the definition of a white point and the use of chromatic adaptation functions to convert colors so they appear consistent, relative to each other, when the observer is adapted to different white points.  Lab is the beginning of an appearance model, but it assumes full adaptation to the white point and doesn't take into account appearance changes due to other factors such as background, surround, and luminance levels.

CIELab is just a math transform of XYZ for a specific white point. L*a*b* values are an invertible function for any given X,Y, and Z operating with a specified white point. It does not include adaption but does try to achieve a more perceptually uniform space than XYZ.

The XYZ/Lab models work well for image capture and reproduction onto reflective media, but when we have a reflective viewing booth side by side with a self luminous display, we experience a breakdown in these models. 
We can strive to get measured XYZ, white point, and Lab values that match between the two, but even when the measured values do match, we don't actually perceive a match.  This lack of a perceived match seems to go beyond what can be explained by instrument error or differences in individual observer.

The breakdown is fundamental. The XYZ values from standard observer spectral functions are averages taken from a 1931, 2 degree, patch matching study over a fairly small group of people.  People vary a surprisingly large amount. You can pick two pure wavelengths that are opposite each other on the xy chromaticity chart and draw a line between them that crosses D50. If you adjust their relative amplitudes such that the resulting XYZ is exactly the XYZ of a D50 spectrum, a "standard observer" would see both as identical whites.  Actual people are not "standard observers" and do not. One person may see a slight yellowish tint, other might see a bluish tint. People's color sense to achieve a metameric match clusters around the standard observer and the cluster locations even vary somewhat between males and females. This is a fundamental issue not addressable by higher precision or more accurate spectrophotometers. And this is why people can set their monitor at exactly the same xy point that their hard proofing setup has but may still see some difference in tint. It's also why LED monitors can appear to have different whites even if their xy coordinates have been set to the same values. This means adjusting the xy on a monitor so their is no tint difference to the viewing illuminant may produce a match for person A but it might not be a match for person B.

In an effort to use the CIE XYZ/Lab models, we adjust the luminance of the viewing booth and/or the display to get a match while eyeballing changes to the image on the display until we perceive a white point match.  Then we build calibration curves for those target values and create a profile for converting an image into display space.  We can get a very close match between viewing booth and display using this technique, but that match is built on the eyeballing, so there's not really an accurate perceptual appearance model underpinning the effort.

CIE CAM02 takes into account levels of luminance, background, surround, and level of adaptation.  Given a paper white background in a viewing booth, a custom illumiinant printer profile for that paper under those lights, and a display that's been adjusted, calibrated and profiled for a white point that matches the measured Yxy values of the viewing booth lights reflecting off that paper white, can CAM02 be used to create a perceived match between print and display without any eyeballing adjustments required?

CIECAM02 is an attempt to predict how colors will appear under different conditions of luminance, surround, and ambience. These change how colors look relative to each other. If you have the same luminance and surround on your monitor and hard proof station CIECAM02 does not apply.

[BradFunkhouser]

So the model works better with reflective media than with self luminous displays because the spectral power distributions of the paper/ink/illuminant colors are more complex and smooth compared to the spikes of RGB in a display, and individuals are extra sensitive to differences in white points, so that's where the breakdown due to the spiky nature of displays and individual variability becomes most apparent.

But could there be more to the softproof mismatch than individual variability?   Here's an article that references a Rochester Institute of Technology professor's description of an incomplete adaptation issue in our perception of self luminous displays.  The comments below the article provide additional details:

        http://www.color-image.com/2012/02/monitor-calibration-d65-white-point-soft-proofing/

That's why I was trying to figure out if CAM02 might address a partial adaptation mismatch.  ArgyllCMS provides access to conversions using the CAM02  model, but it's not clear to me if that model really addresses the type of partial adaptation described by Dr. Fairchild.  I don't think it does.

Can someone point us to additional research in the area of softproofing mismatch and the possible causes:   individual variability versus incomplete adaptation?

[GWGill]

Can someone point us to additional research in the area of softproofing mismatch and the possible causes:   individual variability versus incomplete adaptation?

I've come across several papers with quite different explanations for softproofing mismatch over the years. Whether there is a such a problem or not itself seems to be uncertain - in my personal experience in setting up such a system some time ago we did not encounter such a problem.

There are several mechanisms that can explain a visual mismatch, and all could contribute in any particular situation. Observer variability is one mechanism that may have been under-appreciated in some of the papers addressing this topic, so it could be that their conclusions are misleading. I know Dr. Fairchild has been researching observer variability over the last few years, so it would be interesting to ask him what he currently thinks about the topic of soft proofing. I'll try and remember to ask him in November if he is at the CIC in San Diego.

[BradFunkhouser]

Thanks Doug for helping me think more clearly about individual variability.  I'd been fretting mostly about instrument limitation, though I'm running my spectrophotometer at 3.33nm spacing, which is the best I can do.

Thanks Graeme, for building ArgyllCMS, and for help with understanding and pointers to research.  I look forward to hearing what Dr. Fairchild has been studying if you get a chance to talk to him.

I've dialed in visual white point matches with four different lighting mixes ranging from D50 to D60.  My eyeballs/brain consistently need the display shifted to higher color temps to perceive a match.  The DeltaEs between viewing booth white and my visual display match white range from 5.2 with D50 lights down to 3.3 with D60 lights.  Does my personal variability include a self luminous incomplete adaptation component?  Don't know.  Interesting stuff.

For fun, I had an alternative observer inspect the D50 viewing booth white side-by-side with my eyeballed display white match.  The obsever was first told explicitely to adapt to 5000k.  Here's what that observer saw:

[BrianWJH]

Observer variability is one mechanism that may have been under-appreciated in some of the papers addressing this topic, so it could be that their conclusions are misleading.

Interestingly, XRite's online color acuity test states that 1 in 12 men  have some form of color vision deficiency whereas 1 in 255 women exhibit some form of color deficiency, so does this mean that if the 1931 testing was based on mainly male participation then there's every possibility that the results are highly variable?

Brian

[Doug Gray]

Interestingly, XRite's online color acuity test states that 1 in 12 men  have some form of color vision deficiency whereas 1 in 255 women exhibit some form of color deficiency, so does this mean that if the 1931 testing was based on mainly male participation then there's every possibility that the results are highly variable?

Brian

The study was done with people that had "normal" color vision. Roughly 8% of males have some degree of color vision impairment, red-green discrimination being the most common. It's a defect on the X chromosome but, since women have two X chromosomes, they have to have the defect in both for this to be expressed.