Big difference between (good) proofing paper and final print paper!

[MHMG]

Sorry, I don't get it. What is this "appearance" then, that is so badly measured by photometry? Doug's commentary makes sense to me, that under specified lighting conditions taking account of possible OBA's and different specular/diffuse ratios, then the numbers tell us the story. Otherwise it's a bit of a philosophy argument: "I agree with science accept when I don't like its conclusions, at which point I choose to believe whatever I feel like."

If "appearance" is so badly measured and apparently important, why should anyone give a damn about screen and print calibration?

The CIE color model and and ICC-profiled workflow, albeit not perfect, does a pretty good job at mapping source colors and tones to output color and tones under a wide range of output media dependent conditions, but that mapping is not an exact science and is further complicated by the way the color measurements are made (instrument optics, choice of illuminant, specific measuring conditions like the M0, M1, M2, and M3 specification, spectral properties of the colorants used to reproduce the colors, etc).

dE (and its various flavors like dE2000, etc) is widely cited, indeed deeply entrenched in the graphic arts industry, as a means to evaluate the accuracy of said color and tone reproduction. dE is certainly a useful tool for process control in the graphic arts industry, paint and textile industries, etc. However, dE a perceptual measure of relatively small errors in lightness, hue, and chroma attributes only. Color for color's sake as it were.  As such dE was intended for use under simple viewing conditions when evaluating the perceived match or lack thereof between two colors sitting side-by-side and otherwise isolated on a neutral surround.  Paintings, drawings, and photographs inject another fundamentally important property, i.e, visual contrast, that is not measured by color difference models.  Relatively large dE errors can in many instances go totally unnoticed or at least can be accepted as satisfactory image reproduction quality by the viewer, but in other images smaller dE errors can cause total havoc on image quality by destroying subtle visual contrast relationships.  Moreover, dE and all of its various flavors including dE2000 don't weight hue and chroma errors the way humans do in the context of scene content interpretation (i.e, is the sky a realistic blue, are skin tones plausible, is scene lighting "cool or warm", etc). For these reasons an algorithm designed specifically as an image appearance metric offers a better assessment of overall color and tone reproduction quality. I wrote about one such image appearance algorithm, the i* metric, here:

http://www.aardenburg-imaging.com/i-metric/

A PDF version of the article can also be downloaded on that web page. There are also other articles about the I* metric on the Aardenburg website including the original technical article that was published and presented at an IS&T conference in late 2004.

cheers,
Mark
http://www.aardenburg-imaging.com

[Mark D Segal]

.................dE was intended for use under simple viewing conditions when evaluating the perceived match or lack thereof between two colors sitting side-by-side and otherwise isolated on a neutral surround.  Paintings, drawings, and photographs inject another fundamentally important property, i.e, visual contrast, that is not measured by color difference models. Relatively large dE errors can in many instances go totally unnoticed or at least can be accepted as satisfactory image reproduction quality by the viewer, but in other images smaller dE errors can cause total havoc on image quality by destroying subtle visual contrast relationships.  Moreover, dE and all of its various flavors including dE2000 don't weight hue and chroma errors the way humans do in the context of scene content interpretation (i.e, is the sky a realistic blue, are skin tones plausible, is scene lighting "cool or warm", etc). For these reasons an algorithm designed specifically as an image appearance metric offers a better assessment of overall color and tone reproduction quality. .............

Yes, this rings true from my experience with these comparisons.

Taking it a bit onward, let me ask you whether the following makes sense in terms of what you are saying about the essential difference between dE and i*. I think here we're distinguishing between "accuracy" and "image appearance quality". dE is a metric of "accuracy" (say between a reference file value and a printed patch of that reference file value) that tells us nothing more than whether the reproduction matches the numbers in the originating file. "Image appearance quality" takes account of how people interpret colour in the context of a multi-coloured scene, and your i*metric is designed to measure the extent of that quality. Now here I'm asking about any relationship between the two: from your experience looking between the two metrics, would it be generally valid to say that on the whole the smaller the dE, the better will be the i* result?

[MHMG]

..Now here I'm asking about any relationship between the two: from your experience looking between the two metrics, would it be generally valid to say that on the whole the smaller the dE, the better will be the i* result?

It's a bit like the broken clock story. The broken clock face tells the time accurately two times a day At other times, the error compared to a properly working clock are profound. So it is with dE versus the I* metric. Like the broken clock dE correlates uniquely with I* when it is a perfect 0, and I* color and tone reads a perfect 100% in that situation. Hence, like the two clocks they do begin to correlate under some circumstances, but in other situations, they tell completely different stories. I* metric reveals a correct and informative answer when dE can be profoundly wrong.

Try this simple exercise in PS, and I think you will understand more clearly why dE measurement statistics averaged over many image patch colors doesn't necessarily represent the potential tone and color reproduction problems one sees in a real image. Open a new document in PS, fill the background with L-50, a=0,b=0. Now add a text layer in with easy-to-read font size on top with color picker tool set to write the text color at L=52, a=0, b=0. Then add another text layer with text color set to L=48, a=0, b=0.  Because, a* and b* are 0 in all three colors defining this "image", we have a very simple image containing image information content spatially recognized as the text phrases you typed in the image.  The lighter text and the darker text are more or less equally discernible because of the L* contrast ratio (and dE as well) between the two text colors and the background color are the same for both sets of text.  The dE between both text colors and background is 2.0. Not a very big difference, but enough to create real subject contrast and subsequent spatial recognition. Now fill the background layer with L=51, a=0, b=0 solid fill color. This is also a simple dE move of 1.0, i.e., a "just noticeable" difference with respect to the original background color, but it now impairs the human observer's ability to read the lighter text while at the same time increasing his/her ability to discern the other remaining (darker) text layer. I* tone would return a 50% accuracy value for the lighter text versus background contrast shift and about a 75% accuracy value for the darker text versus background contrast shift. I* tone thus correctly identifies that the visual changes affect the perceived visual contrast of the lighter text image area worse than the darker text image area, and that this is represent a very significant change in overall tone reproduction characteristics of this simple image. Meanwhile, the dE errors would be 0 for the lighter text (it didn't change compared to its starting value) 0 for the darker text (it didn't change compared to its starting value) and a very modest 1.0dE for the background fill color change. The I* metric tells a much more truthful story because it is evaluating image contrast. dE gets it wrong because it's not tracking the image contrast.

I kept this exercise simple using just shades of gray, but you can imagine that when lightness and contrast relationships are made using different colorants, then even relatively small errors in some colors will have bigger impact than the same dE errors in other colors depending on any key information content they are providing to any specific image.

A similar situation occurs for color accuracy weighting issues that dE doesn't get right, but I* Color does get right.  Examples can be found in the article I cited.

cheers,
Mark
http://www.aardenburg-imaging.com

[Mark D Segal]

Thanks Mark, I'll need to set some quality time aside to look into this and your article.

[Ernst Dinkla]

The OP, Narikin, started this thread on the subject of making test prints on a paper that has a lot of resemblance to the paper for the final print. Both printed on the same inkjet printer, white point close, satin like surface. His comment on the difference;

>>(and nb it was not a color shift between the two prints, they were both bang-on,  but in the pricey paper there was more color-depth overall. As Mark says you would not have appreciated it till they were side by side, then it was immediately obvious) <<

With the conditions sketched I still think there is a gamut difference between the two papers. It is also unsure whether the paper profile was up to date for the Canson Platine 2017 version. The image Narikin used + rendering choices showed these differences I estimate.

Tony on the other hand had two papers that resembled one another a bit less, the proofing paper having slightly more OBA, the Epson Platine without it. Surface similar to the OP's papers. Tony's printer profiles were fresh for both papers and the image an sRGB assigned one perceptually rendered in printing. His comment;

>> The printed results on the two papers were almost identical visually, and the gamut maps of the two profiles were almost identical as well. We also tried the canned Epson profile for the Platine and it was close to our created profiles but not as exact a match as printing from the profiles that we created. Printing was done on an Epson SureColor P9000. <<

I do not see much reason to go the 'color appearance' route for explaining the visual difference in the OP prints. Based on the comments of both Narikin and Tony and having samples of the 4 papers here.

The , 'it is a proofing paper' does not cut wood either (to use a Dutch expression), the paper Narikin used has a very good gamut and more than enough visual similarity to the Epson  or Canson Platine. Possibly not enough gamut for the last version of the Canson Platine when the image used, and rendering selected can benefit more of the better gamut.

Way more difficult to create a visual match is when you get Epinal prints of more than a 100 years old, nicely discolored due to the paper making processes of that period and all with different textures. Inks not the same and in the originals mainly spot colors. Sure then in the end you have to create an appearance match with the right paper texture selected etc. Select color adjustments as a last step.

Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
March 2017 update, 750+ inkjet media white spectral plots

[narikin]

The Platine plots show the increasing white reflectance between old and new batches of the Canon quality but also those of other brands. I am still convinced the source is the same.

For some reason I didn't get any notifications of new posts recently, so am pleased to see this thread has found legs!

Ernst - thanks for the amazing screenshot of Platine's improved white point history. It seems you've added a new sample of 2017 Canson Platine recently, and its great to see that back at the front of things. It's like watching a horse race in slow motion as they increase the white point of this all cotton OBA free paper, between the different iterations of it. Quite remarkable where it is at now!!

As an aside, I'm not surprised this thread has sparked discussion - it should concern all who strive to print well on expensive paper. We can't use the $$$ stuff for all our trial/test prints, at least ordinary mortals can't, so some kind of affordable accurate proofing paper is required. I believed that was Epson's Proofing Semi Matte. In fact it kind of still is, as it gives me a very good representation of my final images, on OBA free base, but switching to Platine for the final print is like hitting Turbo. (very unscientific terminology, I know!) That jump is perfect in one way - you truly feel you are getting your money's worth! The 60x80" Platine outputs cost about $80 each in paper alone, if you factor in a some nozzle checks and leading/trailing paper tongues to roll the final print safely. So I can't afford to have 3-4 goes on each image before getting it right.

Few details: the image is in ProPhoto, from a Phase One IQ3-100 back, processed in Capture One, in 16bit. So it has a good large color space, probably exceeding ARGB in a few places, and definitely not fitting sRGB! The Canson Platine was Q3/17. Manufacture date I'm not sure of. EPSWM likewise. Profiles are a mix - I've my own from iSiS and the OEM ones. It didn't change things that much.

Thanks all for informative contributions. We nearly all need to proof on something, so getting to the bottom of the best way/materials to do this is helpful.

[Doug Gray]

It's a bit like the broken clock story. The broken clock face tells the time accurately two times a day At other times, the error compared to a properly working clock are profound. So it is with dE versus the I* metric. Like the broken clock dE correlates uniquely with I* when it is a perfect 0, and I* color and tone reads a perfect 100% in that situation. Hence, like the two clocks they do begin to correlate under some circumstances, but in other situations, they tell completely different stories. I* metric reveals a correct and informative answer when dE can be profoundly wrong.

Try this simple exercise in PS, and I think you will understand more clearly why dE measurement statistics averaged over many image patch colors doesn't necessarily represent the potential tone and color reproduction problems one sees in a real image. Open a new document in PS, fill the background with L-50, a=0,b=0. Now add a text layer in with easy-to-read font size on top with color picker tool set to write the text color at L=52, a=0, b=0. Then add another text layer with text color set to L=48, a=0, b=0.  Because, a* and b* are 0 in all three colors defining this "image", we have a very simple image containing image information content spatially recognized as the text phrases you typed in the image.  The lighter text and the darker text are more or less equally discernible because of the L* contrast ratio (and dE as well) between the two text colors and the background color are the same for both sets of text.  The dE between both text colors and background is 2.0. Not a very big difference, but enough to create real subject contrast and subsequent spatial recognition. Now fill the background layer with L=51, a=0, b=0 solid fill color. This is also a simple dE move of 1.0, i.e., a "just noticeable" difference with respect to the original background color, but it now impairs the human observer's ability to read the lighter text while at the same time increasing his/her ability to discern the other remaining (darker) text layer. I* tone would return a 50% accuracy value for the lighter text versus background contrast shift and about a 75% accuracy value for the darker text versus background contrast shift. I* tone thus correctly identifies that the visual changes affect the perceived visual contrast of the lighter text image area worse than the darker text image area, and that this is represent a very significant change in overall tone reproduction characteristics of this simple image. Meanwhile, the dE errors would be 0 for the lighter text (it didn't change compared to its starting value) 0 for the darker text (it didn't change compared to its starting value) and a very modest 2.0dE for the background fill color change. The I* metric tells a much more truthful story because it is evaluating image contrast. dE gets it wrong because it's not tracking the image contrast.

I kept this exercise simple using just shades of gray, but you can imagine that when lightness and contrast relationships are made using different colorants, then even relatively small errors in some colors will have bigger impact than the same dE errors in other colors depending on any key information content they are providing to any specific image.

A similar situation occurs for color accuracy weighting issues that dE doesn't get right, but I* Color does get right.  Examples can be found in the article I cited.

cheers,
Mark
http://www.aardenburg-imaging.com

Mark, the example background change is 1, not 2 dE. (50 -> 51).

Your point in the I* article is that local variation, especially contrast and hue, over short distances between a reference and test image can produce significant deterioration in image quality not captured by dE.  I completely concur.

One of the attributes of a good prints is consistency and rendering smoothness. Consistency is well captured by standard profile generation but local smoothness is not. I've long wished for a good metric that addresses that and your work on I* as a metric for image deterioration over time from colorimetric changes in ink from light and environmental effects has me thinking about possible ways to overcome the intrinsic limit a relatively small set of patches provides in establishing a metric for printer quality. dE obviously correlates but there can be significant variations in smoothness not captured in a dE patch set verification.

What may work quite well for measuring "smoothness" is creating a set of Lab circles of constant L* in steps of dL= 1.43 and print them using colorimetric intent. Using the "quality" option in I1Profiler generates 36x36x36 cubes. Each cube spans 2.9 dL and 7.3 da and db. So this covers the cubes vertices as well as midpoints of the profile cubes along L*. The a* and b* would be a 2 dim. surface where each point is the interpolated values from the profile.

But then what?

A reasonably modern camera can capture the RGB values with high resolution. The images can be processed in linear space using dcraw or equivalent into Lab estimates. The problem with a camera capture is that its not going to be all that accurate in the specific Lab value of some point on the ab circle. But it should provide excellent difference values. One could then create a smoothness metric from the deviations.

It's also possible after identifying the areas of largest deviations to create a special patch set in printer space to explore the nature of the deviations more precisely.

This may dovetail into your area of interest, characterizing long term stability of printer image quality.

[MHMG]

Mark, the example background change is 1, not 2 dE. (50 -> 51).

Please reread my earlier comment.  I do think I said it correctly. The two text layers in the example start with an initial dE of 2.0 when compared "side-by-side" to the background fill color. This simple "image" is also adhering quite well to one of the main tenants of color difference models, namely color perception at 2 or 10 degree subtended angle of view for colors existing side-by-side (when the text size is picked to subtend that sort of dimensional point of view) and then surrounded by neutral gray in the overall visual field of view. When the background fill color is then moved up from 50 to 51 without altering a* and b* values, I stated that this was a dE =1.0 as you also note, but this is now comparing this one solid-fill background color in a "before and after" dE evaluation against itself not to the other two text colors in my example. This procedure is the customary methodology for using dE measurements to track "before and after" image reproduction results, i.e., dE is rarely if ever used to compare near neighbor color value relationships in graphic arts (as I* tone does). dE is used just to track what one single color element is doing throughout the image reproduction cycle. Yet by moving the background fill color just 1 JND (just noticeable difference) value, the lighter text color's visual contrast signal drops 50% to the human observer in this example, and the I* metric flags that change by an L* derived gamma function for near neighbor contrast relationships. That's a huge visual change in image appearance, yet we are only talking about small movement in dE relationships between before and after the color value of just one image color. Had I suggested making text colors with extremely different initial L* values from the background color L* value, even if the dE arose from differing a* and b* values, then such small dE changes to any of those colors would have hardly altered the visual appearance of the before and after image adjustments because the visual contrast would be much higher to start with, and yet the dE analysis would remain the same. There's the problem, and that's why a metric that can actually quantify local image contrast reproduction is so important to provide any reasonable figure of merit when trying to quantify image appearance differences in two images that aren't a dead colorimetric match to one another... As in this whole discussion of creating a proof print that will yield a faithful visual rendition of the intended outcome when compared to the print made on the final selected printer/ink/media combination.

Aging studies that involve tracking image quality before and after some environmental event like light exposure is indeed what drove the development of the I* metric initially, but it was derived on a "first principles" color science approach to signal quality retention in photographic images that go through any kind of color and tone reproduction cycle. In that sense, I* is also highly applicable to any graphic arts color and tone reproduction cycle where one is trying to compare a reference image (e.g., a digital file with encoded LAB color values or embedded ICC profile to derive those values, an electronic display with measured colorimetric values, an original color or B&W print with measurable color values, etc) to a comparison image (e.g a fresh reflection print copy, transmissive film, or another image representation on a different electronic display) in other words, made by any kind of reproduction process. The I* metric just needs to have a colorimetrically defined reference image, and then any other image reproduction you choose as a "comparison" image can be evaluated by the I* metric as long as you can align the color elements (e.g. pixels or color patches) in the reference image with those in the image to be compared.

I'm somewhat surprised no experts in the graphic arts industry have picked up on the I* metric after over a decade since it was published, but as I noted before dE models are so entrenched in our industry that few experts seem to have an open mind to explore any other approaches to tone reproduction quality metrics. Around the year 2000 I began research on what ultimately became the I* metric after nearly five years of study because I quickly determined with simple Photoshop exercises that dE was an epic fail when trying to fairly rate different systems for loss of image quality as they were subjected to any kind of color and tone reproduction errors (mine environmentally induced), but I have always believed the I* metric  is highly relevant to any image color and tone reproduction quality evaluations which are quite common throughout the graphic arts industry, not just the niche arena of environmentally induced print fading and discoloration.

cheers,
Mark
http://www.aardenburg-imaging.com

[Doug Gray]

Hi Mark,

I was specifically referring to this from your post:

Meanwhile, the dE errors would be 0 for the lighter text (it didn't change compared to its starting value) 0 for the darker text (it didn't change compared to its starting value) and a very modest 2.0dE for the background fill color change.

It was obvious from your earlier description you meant 1.0dE, the change in background fill you had earlier described.

As for the experiment, I did it and the results were exactly what I expected. It shows how strong an effect a selective dE change in sensitive regions can produce pretty large perceptual effects. I, for one, would like to see further development of metrics designed to measure a kind of perceived accuracy. No idea why I* or something similar hasn't received more attention. It could be quite useful.

[MHMG]

Hi Mark,

I was specifically referring to this from your post:

It was obvious from your earlier description you meant 1.0dE, the change in background fill you had earlier described.

As for the experiment, I did it and the results were exactly what I expected. It shows how strong an effect a selective dE change in sensitive regions can produce pretty large perceptual effects. I, for one, would like to see further development of metrics designed to measure a kind of perceived accuracy. No idea why I* or something similar hasn't received more attention. It could be quite useful.

Thanks Doug, you were right. I said it correctly in the first part of my comments but made the error you noted in the later part of my comments. I have just now corrected the error in the original post. Again, thanks.

Bottom line: if we want to objectively rather than subjectively quantify human perception of image quality when evaluating a reference image with respect to a subsequent reproduction of that image (which is supposed to ideally be a perfect colorimetric match) then the quantifying mathematics must include a calculation of visual contrast, and the metric also needs to weight color errors with respect to how vivid the colors in the image actually are, human observers being much more sensitive to changes in low chroma colors compared to high chroma colors when trying to interpret the meaning of the color information content provided by the various colors in the scene.  The I* metric does both, dE and all of it's various versions including dE2000 does neither.

The inherent discussion of this thread, i.e., the comparison of a proof print to the final print made on a different media (or any other printing process) is a perfect application for the I* metric. Just sayin... Viewers are hard pressed to discern notable differences between two prints when the I* color and tones scores achieve 90% or better. As I* falls to lower values visual discrepancies between the two prints become more and more noticeable. What dE is reporting under the same circumstances is anybody's guess. The dE measured values could be small or they could be rather large depending on the image information content presented in the original image.

all the best,
Mark
http://www.aardenburg-imaging.com

[Ernst Dinkla]

The inherent discussion of this thread, i.e., the comparison of a proof print to the final print made on a different media (or any other printing process) is a perfect application for the I* metric. Just sayin... Viewers are hard pressed to discern notable differences between two prints when the I* color and tones scores achieve 90% or better. As I* falls to lower values visual discrepancies between the two prints become more and more noticeable. What dE is reporting under the same circumstances is anybody's guess. The dE measured values could be small or they could be rather large depending on the image information content presented in the original image.

all the best,
Mark
http://www.aardenburg-imaging.com

Mark,

I see many cases where the I* metric would increase the fidelity between original and print and in prepress proofing where there are repeat runs for same condition papers + inks it would be applicable too. That is also a field that has the best conditions for it; proofing papers that already aim at the representation of other printing processes papers and CM/prepress methods to bridge the ink/printing differences. In cases where older originals have to be reproduced in inkjet printing it becomes far more complex with lots of unknowns. They usually do not not come with spectral data on the old inks/paints/crayons used. Where reproduction systems start with multi spectral takes of the original there is probably money and methodology enough to add the I* metric system as well.

The thread started on the fidelity between inkjet prints where just one aspect differs in their production, the papers, so even less to bridge as in prepress proofing. Printer, inks, application to print from etc all alike. The papers' gamuts not that far apart either, the surfaces and paper whites sensibly chosen for the task. Of the two cases reported it looks like the OP's case had a few flaws in profiling, the test paper a somewhat smaller gamut and a more demanding digital image to start with. To me improving the CM (custom profiles for both papers) to do the job would be the way to go and adopting prepress CM methods like device link profiling as a more pro step. Where the image claims the gamut of the print run paper totally, the smaller gamut test paper will never get to that level. CM runs out of possibilities then, I doubt I* metric will be of much help than either, not more than what a skilled observer could do to some hues that unbalance the test print compared to the wider gamut end print. It will never get the overall color strength of the final print. Adding ink lay down in the paper preset data for the test paper usually is not working either, compromising on dot gain not the limitation as it is mainly the color fidelity that is sought in test printing, but with pigment inks there is a limit on the max chroma obtained with ink layer thickness increases. Same with dmax.

This thread went another course and that has been interesting too but I fear that the sensible approach of the OP and other contributors to find ways of using less expensive test print methods is not served well by the direction this thread took. I see explications enough on the web how to aim Photoshop's soft proofing to output devices but there is way less explained how Photoshop could help in test print simulation.

Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
March 2017 update, 750+ inkjet media white spectral plots

[Pete Berry]

Mark,

(snip..)
The thread started on the fidelity between inkjet prints where just one aspect differs in their production, the papers, so even less to bridge as in prepress proofing. Printer, inks, application to print from etc all alike. The papers' gamuts not that far apart either, the surfaces and paper whites sensibly chosen for the task. Of the two cases reported it looks like the OP's case had a few flaws in profiling, the test paper a somewhat smaller gamut and a more demanding digital image to start with. To me improving the CM (custom profiles for both papers) to do the job would be the way to go and adopting prepress CM methods like device link profiling as a more pro step. Where the image claims the gamut of the print run paper totally, the smaller gamut test paper will never get to that level. CM runs out of possibilities then, I doubt I* metric will be of much help than either, not more than what a skilled observer could do to some hues that unbalance the test print compared to the wider gamut end print. It will never get the overall color strength of the final print. Adding ink lay down in the paper preset data for the test paper usually is not working either, compromising on dot gain not the limitation as it is mainly the color fidelity that is sought in test printing, but with pigment inks there is a limit on the max chroma obtained with ink layer thickness increases. Same with dmax.

This thread went another course and that has been interesting too but I fear that the sensible approach of the OP and other contributors to find ways of using less expensive test print methods is not served well by the direction this thread took. I see explications enough on the web how to aim Photoshop's soft proofing to output devices but there is way less explained how Photoshop could help in test print simulation.

Met vriendelijke groet, Ernst

I found years ago that papers in the PK class are much more similar than they are different when inked identically, with same media choice, and printed with either printer or PS with identical profile controlling colors. To whit: back in the early days of Baryta papers and my iPF5000, I ran a little experiment with sample sheets of PK, using  Canon's "Special #5" media type with printer control.

Using Digidog's original printer test image, I tested ten different PK papers, including RR's Polar Satin and Ultrapro Satin-II as proofing candidates, with Ilford GFS, Harmon FB Al Gloss, Hahne. FAP and PRP as archival baryta possibilities. The 8x10's were then scanned in my V700 with identical settings, including white point set to show base paper differences. Results of six below:

https://www.flickr.com/photos/8075035@N03/albums/72157624569134778

While there are certainly differences, with the Harmon FB Al Gloss standing out with a slight yellowish cast, and the RR Polar Satin the coolest as well as slightly flatter overall, scattered on a table they look remarkable similar. I ended up with the RR UPS-II as proofer, though slightly warmer, and the GFS as display papers. For a very good match I knock down yellow just a bit (1/30) in my PS 16-bit Plugin's saved proof settings, using Special #5 media type, and the same generic Canon 5100 paper profile for both - Canon Photo Paper Plus semi-gloss-Highest - far better than the Ilford profile for GFS, and a custom GFS profile in the past.

So my suggestion to the long lost OP is to first simply try a proofing paper print using the display paper's profile, and tweak it from there if needed. I'd be interested in other's results if they did the same. A bit of heresy in this uber-technical discussion, but after all, it's the best match to the display paper, not the best free-standing proof print we're looking for here...
Pete