Fancy Graphics Galore
Some observations on Gamutvision and PrintFIX Pro
By Gerard Kingma
Boy, Norman Koren certainly has a helluva lot more brain cells in his mathematical drawer than I do. Recently I bought Mr. Koren's Gamutvision because I wanted to find out whether I'd done right in buying PrintFIX Pro to make profiles for my HP Designjet 130. I saw more mathematical wizardry than my linguistically oriented brain could handle.
Even though I don't understand half of the math in Mr. Koren's instructive manual for Gamutvision, I do have a basic grasp of the underlying principles and even though I am on a budget, I do take color management very seriously. Otherwise, what would be the point of shooting 4 x 5" on Velvia, scanning the plates on a Polaroid SprintScan 45U, post-processing the scans in Adobe Photoshop CS2 and sending them to a HP Designjet 130 in order to make 24" x 28" fine-art prints, as I do?
When I bought PrintFIX Pro, my initial thoughts were moderately positive. In the real-world print that I chose to test at the time, the end result showed a slight improvement over the print that I produced with the standard HP profiles. Now I'm not so sure.
Input: From the Polaroid SprintScan 45U to AdobeRGB(1998)
Gamutvision turns out to be a superb tool to help compare various profiles, and it has taught me a lot over the past few days. Much of the time I just had plain fun producing really spacey graphics that looked way cool, without having the slightest idea what I was looking at. But as I was eager to find out if my low-budget version of a color-managed workflow could actually stand the test of objective inspection, I pressed on and started at the input side, my film scanner. I had previously calibrated my Polaroid SprintScan 45U with a dedicated target for the new Velvia 100-film, produced by the ever helpful Wolfgang Faust. This is a screenshot from Gamutvision that shows the profile for the scanner, produced with LCMS, together with AdobeRGB(1998). It's just the one gamut overlaid on the other gamut, without mapping the colors from the one gamut to the other. As you can see, the scanner's gamut (wire) is much larger than AdobeRGB (solid). AdobeRGB contains some lighter blues and cyans that apparently the scanner doesn't pick up, or the film can't record; the AdobeRGB-gamut sticks out slightly at the top of the scanner gamut.
[From scanner to workspace, no mapping]
But in a real world scan, you'd map - or render - the colors from the scanner's gamut to the workspace, i.e. AdobeRGB. For rendering intent, the obvious choice would be Perceptual or Relative Colorimetric, the latter with black point compensation. If you do that, a curious gap or black spike emerges in the deep blues:
[From scanner to workspace, mapped]
The way I interpret this, is that this particular combination of film and scanner (Velvia 100 and Polaroid SprintScan 45U) can't record very deep blues. I'm open to other interpretations and comments on the cause of this spike. As most of these very deep blues appear to be outside of AdobeRGB anyway, I doubt that this will present problems in real-world prints. Otherwise, the scanner profile looks perfectly usable. Mr. Koren kindly took the time to have a look at this profile, and this is his take on what we're seeing here: Gamutvision uses a test pattern with the most saturated possible RGB values, but the scanner may not ever output these values in practice. So the actual scanner gamut may be smaller than the gamut displayed by Gamutvision. No easy way around this, though the new Imatest Multicharts module will have some useful results for the IT8.7 chart. However, the areas where AdobeRGB has a larger gamut than the scanner are valid. When you reduce the saturation to 0.8, the "gap" or "black spike" disappears. So I'm pretty sure it represents RGB values that the scanner would never output in practice, and hence the profile-making software didn't deal with.
Output: From AdobeRGB to the HP Designjet 130
So, on to the output side. I was eager to see how the profiles I had produced with PrintFIX Pro for the HP Premium Plus Photo Satin paper that I use on the Designjet 130 would compare to the standard profiles supplied by HP for this (excellent) paper. There are four profiles to consider: two for HP, quality settings Best and Max detail, and two for PrintFIX Pro, also quality settings Best and Max detail, all mapped with four rendering intents: Saturation, Perceptual, and Relative Colorimetric with and without black point compensation. Absolute Colorimetric isn't suited to real-world images, and black point compensation only affects Relative Colorimetric rendering. This gives 16 permutations, so I printed the test image supplied with Gamutvision for each permutation:
[16 test prints]
The screenshot below from Gamutvision shows the workspace (AdobeRGB) mapped to the gamut for HP Photo Satin according to the PrintFIX Pro-profile, quality setting Max detail, rendering intent Relative Colorimetric with black point compensation. The gamut for the PrintFIX Pro-profile appears to be larger than the HP-profile, but most noticeable is a fairly strong banding in the darker greens for the PrintFIX Pro-profiles.
[From AdobeRGB to PFP, Relative Colorimetric, Maxdetail, screenshot]
The HP profiles are much smoother in this area.
[From AdobeRGB to HP, Relative Colorimetric, Maxdetail, screenshot]
The L*(output)-view, below, looks more like a ski slope than a balloon - particularly suited to studying this kind of behavior:
[From AdobeRGB to PFP, Relative Colorimetric, Maxdetail, screenshot: L*(output)]
Mr. Koren commented: Most of the time these ripples have no visible effect, but they may occasionally cause problems similar to the "smudged pine" problem I discuss on http://www.gamutvision.com/docs/smudged_pines.html
Good news from a scientific point of view: this shows up equally well in the actual prints, as reproduced below, photographed with a digital camera. Obviously, draw no conclusions for actual colors from what you see here on your monitor, but go by my observations.
[PFP, Relative Colorimetric, Maxdetail, actual print]
[HP, Relative Colorimetric, Maxdetail, actual print]
PrintFIX Pro recommends using Saturation rendering intent rather than Perceptual or Relative Colorimetric. While in theory this wouldn't be the obvious choice, in practice it's not at all a bad recommendation, as the Saturation intent produces a much smoother result:
[PFP, Saturation, Maxdetail, actual print]
However, the medium greens in this setting show a significant darkening, as you can easily see on the actual print above. The HP actual print in this area is much closer to the original test image.
So what does this mean? In real world images, if you were to print a picture with smooth transitions in dark green areas, you'd expect banding with the PFP profiles in Relative Colorimetric rendering intent. Also, medium greens would not be reproduced very accurately. For a real-world image, I chose a scanned 4x5"-shot of a fern I recently photographed in the Vosges, eastern France. The fern leaves contain a myriad of greens, and that's what makes this a nice shot printed at 24" x 28".
[Velvia scanned for web presentation]
[ferns printed on A3+, with 16 test prints]
As I expected from these tests, the medium greens on the fern leaves are represented with more vibrant greens and more subtle transitions in the fern leaves through the HP profile, rendering intent Relative Colorimetric with black point compensation (the best HP result) when compared to the PrintFIX Pro profile, rendering intent Saturation (the best result for PFP), both with the setting Max Detail in the HP driver. There's no way of showing this with any degree of accuracy on a website, but trust me on this: the simulation with Gamutvision is confirmed by the real-world print.
The verdict on Gamutvision
This is an excellent tool, extremely informative. In order to understand what you see on screen, you should definitely take the time to read the excellent online documentation and other articles about gamuts, rendering intents and whatnot, otherwise it's just fancy graphics. I wrote most of this article with version 1.1., which was slightly buggy, it crashed a couple of times. Version 1.1.9 has just been released; Mr. Koren replied, As for the bugs, we try our best to fix any that appear. Some may be related to the OpenGL renderer used for the 3D L*a*b* plot. We frequently update Gamutvision, as we come up with new ideas. Highly recommended.
The verdict on the HP Designjet 130
I've been using this printer for more than a year now and it has worked splendidly for me. I've had two solo exhibitions with 24"-prints from this machine, which have elicited many oohs and aahs from visitors. When I look at this latest print of the ferns, in sheer print quality there's really very little left to desire. Highly recommended, especially if one considers its price point.
The verdict on PrintFIX Pro
I produced the PrintFIX Pro-profiles from the most detailed targets, with 729 color patches to read. This takes a while, but it's doable. However, the banding in the darker greens suggests to me that measurement targets with even more patches might be called for in order to produce smooth transitions. This may be true only for this particular printer and paper combination, I don't know. But I'd like to see a next software version with the option to produce targets with more color patches, and see if that would make a difference. I asked ColorVision/Datacolor, creators of PrintFIX Pro, to comment on this suggestion. Mr. C. David Tobie, Product Technology Manager of ColorVision, informs me that more patches won't necessarily make the profiles smoother: More patches may well build more precise profiles, but they may, in that very precision, be less smooth. Its something of a tradeoff between literalism and smoothness.
As it is, the standard profiles from HP produce better results. If I'd known that, I wouldn't have bought PrintFIX Pro. But that doesn't mean that the package produces poor profiles; they are perfectly usable on real-world prints. So if you have a printer-paper combination for which you cannot download good profiles, then you might want to consider purchasing PrintFIX Pro; it's certainly a lot cheaper than many other spectro-based printer profiling solutions. If you use only one or two printer-paper combinations without profiles, you might want to consider ordering third-party custom-made profiles from other sources. And of course, with the introduction of the new Z-series printers, HP has shown the way into the future with built-in color management features that will make third-party profiling solutions more or less obsolete. With a Z-series printer in the office, I would have a lot of fun with Gamutvision because it would enable me to compare various profiles for numerous (third-party) papers, produced by the printer itself!
I offered ColorVision/Datacolor the opportunity to respond to this article prior to its publication. Mr. Tobie sent me an interesting reply:
I would note that default profiles shipped with printers are often manipulated in ways that offer advantages for showing vector gradients, such as the rainbow gradient you are using, over custom profiles which are built directly from measurements. Your statement that the HP canned results are "more accurate" on certain greens is not as simple as it sounds: out of gamut colors may be brought into gamut in a number of ways, and which way appears to most closely match the unobtainable original color is somewhat subjective: match the hue, the saturation, the brightness, some prioritized mix of the three?
The type of banding you note in the outer reaches of the gamut is not uncommon in patch-built profiles, in fact I have seen it in profiles from every major manufacturer of profiling software; but you are not comparing PrintFIX PRO to other profile building software, you are comparing it to manipulated OEM profiles, which is more of an apples to oranges situation. Its worth pointing out that these vector color regions at the gamut edge are seldom if ever reached in actual photographic content.
Indeed, I would be the first to agree that it is easy to make observations from theoretical visual representations such as these that have little bearing, if any, to real-world prints, and that many real or imaginary shortcomings of these profiles will rarely show up in real-world prints. That's why I will go on making test prints from real images; and that is what I base my final judgment on - practice, not theory. I don't quite know what to make of Mr. Tobie's suggestion that HP are actually doctoring their profiles to make them look good in vector gradient representations such as these, but there it is.
I once more consulted with Mr. Koren, and he replied: Debate is a good thing, and Mr. Tobie's comments are well worth publishing, but I can't completely agree with them. For one thing, you can run the "ski-slope" L* plots (or any of the Delta-L* or Delta-E* plots) for less saturated colors. The bumps go away slowly. They are present for in-gamut colors. I seriously doubt that the HP profiles are manipulated for unfair advantage. Not that many people run rainbow-gradient tests (thus far, anyway). It is entirely fair to compare manufacturer-supplied profiles with the results of profile-building software because they have exactly the same function. It's not apples-oranges.
But it is true that colors at the gamut limits are rarely encountered. Most profile defects show up rarely - and only on images that contain colors in the troubled region. The purpose of a gamut viewer such as Gamutvision is to preview what a profile can do so you don't have unpleasant surprises and so you understand the limits of your printer/profile combination.
If Mr. Tobie is right and HP are optimizing their profiles for maximum performance in printing vector based colors such as produced by Adobe Illustrator (because that's what he means with his remark), wouldn't these profiles automatically also be optimized for real-world photos? Surely, the smoothest possible profile would be the most ideal profile for any type of image? Shouldn't everybody be busy optimizing their profiles for smooth rendition of gradients?
I bounced this idea off Mr. Koren, and he replied: I think that smooth gradients are extremely important - we often see them in real situations, like skies. So the answer to your question would be, yes. A profile should handle everything sent to it gracefully. Of course there are four rendering intents to cover a variety of applications: vector gradients might typically be used with Saturation intent. I'm finding that I usually prefer Colorimetric intent (with Black Point Compensation) to Perceptual intent. Perceptual rarely does exactly what the textbooks say.
Let's take a closer look at the notion that defects at the outer edges of the gamut are seldom visible in real-world prints. Below you'll find screenshots from yet another powerful function of Gamutvision, the Round Trip view. Basically what you do here is let the profile bounce off of its own mirror image. The returned image should be similar to the input, as it were. Any differences show up very easily. The first is a Round Trip view for the PrintFIX Pro-profile, for quality setting Best in the HP driver:
[Round Trip PFP Best]
The same bumps show up, as well as a curious yellow spike. Again, the HP profile for this setting is much smoother:
[Round Trip HP Best]
But note also that these are not views for 100% Saturation, but for 74%. This means that these defects do not show up only at the outer edges of the gamut, but also for less saturated colors, as Mr. Koren suggested. The Saturation slider puts you in the pilot's seat and lets you fly straight through the cloud to see what's lurking there.
Naturally I was very eager to learn what HP would have to say about all this. As to Mr. Tobie's suggestion that HP's profiles are optimized for vector-based colors rather than real-world photographic image content, the official reply from Johan Lammens, Senior Color Scientist, Hewlett-Packard ICD R&D Department, is brief but very clear: HP provides ICC profiles with its Designjet professional photo printers for a wide range of paper types. These profiles are optimized for best printed results, and not for best display in analytical tools, as Mr. Kingma's results demonstrate very nicely.
Hmmm. It doesn't come as a big surprise to me that HP denies the suggestion that they're optimizing their profiles for use with vector-based colors. Why would they. But when I re-read their statement, it strikes me that there's also something that they're not denying - that the profiles are actually optimized. And that does surprise me. I would have thought that profiling is an objective process - the input values for your prints are known variables, expressed for instance in Lab, RGB or CMYK values. You send these to the printer. You then measure the colors of your output print with a spectrometer, and then you build a profile from those values to correct any differences. And Bob's your uncle. Right?
Well, no. It's not that simple. Why not? Because there are colors you can print with a particular printer on a particular paper, and there are colors you cannot print on this particular paper with this particular printer. What if your image contains colors that this printer-paper combination can't handle? Well, you can throw them away because they're not printable, or you can change them to values that the printer can handle. In other words, you can render them to values that are inside the printer/paper's gamut. That's where rendering intent comes in. And apparently, it's up to the manufacturer to decide how out-of-gamut colors are rendered to printable colors.
Mr. Ignacio Ruiz de Conejo also works with HP's Inkjet Commercial Division in Barcelona, the group that actually develops HP's large format printers and associated profiles. He informs me that there are many ways to implement the theory behind rendering intents in practice, and indeed, many, many roads lead to Rome, if not all. According to Mr. Ruiz de Conejo, The Saturation intent is only loosely defined in the ICC profile specification, so that manufacturers can provide what they consider to be the best implementation of it. It is implied only that the device will try to stretch the device capabilities to its maximum potential, or to "saturate" the colors as much as possible. That seems to be what PrintFIX is doing. Likewise, the Perceptual rendering intent is defined fairly loosely in the ICC profile specification. End users tend to assume that in-gamut colors are preserved exactly and only out-of-gamut colors are remapped in the Perceptual intent. We do allow for some adjustment of the in-gamut colors, however, so that the resulting image is more pleasing, rather than colorimetrically accurate. Hence our Perceptual intent could be considered akin to Saturation in a way, because it does not strictly maintain the colorimetry of all in-gamut colors. But it is not "hand-tweaked for vector colors". We use internally developed profiling software and real (photographic) images rather than synthetic images or vector colors to evaluate image quality for our professional photographic printers.
Your choice of Relative Colorimetric with Black Point Compensation is understandable, since that is what many professional photographers use. But Perceptual intent is, or can be, a better choice. The problem (for the user) is that Perceptual implies you are leaving color mapping decisions to the device or profile manufacturer, and that it may not agree with your personal color preferences. Therefore, Colorimetric with BPC may seem like a more predictable or more reliable choice. What perhaps many users donâ��t realize is that black point compensation, as implemented today by Adobe, moves all colors and not only the out of gamut ones. Perceptual has fewer restrictions, and it is up to the manufacturer to design the color transformation as they see fit. Maximizing the gamut, maintaining gray neutrality, preserving primary hues, minimizing grain, preserving lightness and/or hue for out-of-gamut colors... In the end, the profile manufacturer has to make decisions that may not be to everyoneâ��s liking.
So, there you have it. You can buy a profiling package at over a thousand US dollars, or you can buy PrintFIX Pro at roughly half the price, or you can order custom-made profiles, or you can choose to use the profiles supplied by the printer's or sometimes the paper's manufacturer. Whatever the differences may be, the one solution will in a way not be objectively better than the other solution. Ultimately, the manufacturers decide for you what they think will look best, because they're the ones that take decisions about compromises in the way the profiles are built. In practice, you may very well prefer a print made through a standard profile over a custom-built profile, or the other way around, only to find that next day, you prefer the other solution on a different print. Even then, the differences will probably be small and with a few adjustments, you can probably make the one look very much like the other. As Mr. Tobie of ColorVision puts it, Mr. Ruiz de Conejo feels Perceptual is a better intent choice with HP DesignJet profiles; with ColorVision profiles we often recommend, and many users choose, the Saturation intent, where your issue is much reduced. However, prioritizing saturation over other factors can cause other variations in hue or lightness, such as the darker greens you note. I would suggest softproofing in Photoshop, and either bringing your ferns into gamut using the softproof and gamut warning tools, or lightening the areas that you find too dark. After all, printer profiles do not provide perfect results blindly; they provide the tools to predict and adjust your output, minimizing trial and error, and the waste of time and materials it involves.
Well yes, but that only gets me back to square one - fiddling with adjustments until the print looks like something that I can live with. That kind of random excellence, or lack thereof, is not what we want of course. What we want is printing profiles that will reliably and measurably reproduce colors in print as we see them on screen. From what I've learned from the experts over the past few weeks, it'll get better and better, of course, but it's also very much a quest for the Holy Grail in color management - it may never be found.
Gerard Kingma is a semi-professional landscape photographer based in the north of The Netherlands. Please visit his website at http://www.kingma.nu for more information on fine-art prints, digital orders, exhibitions and workshops.
Gamutvision by Norman Koren: http://www.gamutvision.com
PrintFIX Pro by ColorVision/Datacolor: http://www.colorvision.com
Hp Designjet: http://www.hp.com/go/designjet
Calibration targets by Wolfgang Faust: http://www.targets.coloraid.de
Gerard Kingma Travel and Nature Photography: http://www.kingma.nu