Dan Margulis Applied Color Theory
Gamut Tests of Two Sample Images
RGB Working Space, Part 1 of 2
Posted by: Dan Margulis
Sat Nov 11, 2006 10:33 am (PST)
This list has traditionally taken the position that exotic RGB definitions, those with ultra-wide gamuts or ultra-low gammas, are so unlikely to be used as working spaces by those serious about color correction that extended discussion of them should be considered off-topic.
The broader question of what RGB workspace to use has come up many times on this list and there are several threads have been archived at www.ledet.com/margulis/ACT_postings/ACT.htm. Recently, however, two images have been submitted that again raise the question of whether there can be value to something wider-gamut than most of us use. I have tested them and have uploaded the image files to the Ledet site and analyzed them in the second part of this post. First, it is appropriate to review the basics.
TWO GROUPS, TWO NEEDS
Most list members prepare almost all of their files either for CMYK or for some RGB output where the gamut resembles that of CMYK. Some members, on the other hand, also have to prepare for output devices that have a greater gamut. The two groups have different needs. This half of the posting concentrates on those who do *not* have access to such extended-gamut printers and instead are limited roughly to traditional CMYK.
Narrow-gamut RGBs, by which we almost always mean sRGB although Apple RGB also qualifies, exceed the CMYK gamut in many areas, but in some areas they do not cover all the colors. Moderate-gamut RGBs, like ColorMatch RGB or certain others that have been developed by people who are dissatisfied with sRGB and Adobe RGB, cover almost all of the CMYK gamut, at the price of exceeding it considerably in certain areas. Wide gamut RGBs, of which Adobe RGB is the most prominent example, exceed the CMYK gamut everywhere and exceed it wildly in certain areas. Ultra-wide gamut RGBs like ProPhoto RGB enormously exceed the CMYK gamut everywhere, as does LAB.
For whatever reason, the professional community has split between sRGB and Adobe RGB. Polls at recent Photoshop Worlds suggest that around 95 percent of serious users have chosen one or the other, with the remainder using either moderate-gamut RGBs, sometimes of their own design, or ProPhoto RGB.
Choosing which to use involves several factors, but these posts consider only gamut—the range of colors that each can portray. An RGB gamut that isn't big enough to encompass what the output device can print means that certain colors can never be achieved. A gamut that's too big can result in massive loss of detail when the file is converted into colors that can be printed.
In a perfect world, we would certainly find an RGB gamut that matched the output conditions exactly, neither larger nor smaller. Unfortunately, in our vale of mortal sorrows, all output devices use CMY colorants, which dictate that there is going to be a gross mismatch no matter what RGB we use. The only question is which mismatch does the least damage.
All output devices, presses, inkjet, and otherwise, are relatively poor when producing reds, greens, and blues, because these colors are made up of at least two inks. As inks aren't completely transparent, they can interfere with each other's performance. Solid or nearly solid cyans, magentas, and yellows are the strength areas of all output devices. These are the areas in which narrow- and moderate-gamut RGBs do not cover the entire output gamut.
All output devices also perform relatively poorly as the colors get lighter. This is because they depend on the paper to add lightness, and no paper is perfectly reflective. The worse the paper, the faster the problem occurs. Solid magenta ink on press is outside of the sRGB gamut, but dropping even down to 85% coverage brings it back in. If you're working with extremely white paper, such as many people use on inkjets, possibly a little lower would still stay outside the sRGB gamut.
THE LESSONS OF HISTORY.
As those who follow the history of these conflicts will know, the general pattern is that the group that I sometimes describe as the Conventional Color Management Wisdom lags around five years behind me. Many things *used* to be controversial--whether embedded profiling would ever be a reliable way of passing files to strangers; whether one profiles the proof or the press; whether an inkjet printer can ever be used for a contract proof; whether digital cameras can ever achieve film-like quality; the merits of Photoshop 5; the need for a module to read raw camera data so as to bypass any automated correction; whether a "pushbutton color" situation would ever develop; whether cameras can be profiled for shooting in the field; whether color management cannot succeed without a strong commitment to process control; whether 16-bit color correction creates "a night and day difference, totally obvious to anyone who looks,"; whether conversion to LAB causes "catastrophic damage" to a file; whether printers can be relied upon to convert RGB to CMYK; whether Perceptual or Relative Colorimetric is the right intent for photographic images; whether CMYK images should be tagged; and on and on. In each case, the CCMW argued strenuously against my position before eventually adopting it several years after I did.
In fact, if you go to the ColorSync users list today, and ignore the extremist wing and those who demonize me out of habit, you'll find that it basically occupies the terrain that I have since around 1998, with one notable exception. There is still paranoia about using too small an RGB space.
WHEN THE TARGET IS CMYK OR SIMILAR.
The conventional wisdom has insisted that the RGB definition exceed the gamut of the output device in every respect, forgetting the dangers of *grossly* exceeding it in reds, greens, and blues. People who attend my classes and bring in images that have caused them significant problems. One of the most consistent varieties is detail loss caused by using Adobe RGB on capture when the target is CMYK. One of the most common reader requests for coverage in Professional Photoshop was how to treat RGB and LAB images that contain colors outside of the gamut of the output conditions, and I have page after page of that information in response.
The idea that sRGB could cause problems for CMYK (or photo lab users) is a chimera. As noted above, the only colors that could possibly provoke a problem are heavy, saturated cyans, magentas, and yellows. As a practical matter, the first two don't exist. I have not seen an example in the last five years. Prior to starting work on that section of the book, I asked list members if they had any. Nobody had anything cyan. A couple of people had magenta images (one image of flowers, one of a brilliant silk scarf), and I was already using (in a different context) a picture shot by David Moore of an extremely magenta blouse. All of these were close--but none contained any areas that were out of gamut for sRGB yet printable in CMYK.
As for yellows rich enough to be printable in CMYK, yet out of the sRGB gamut, they exist--yellow peppers, bananas, etc. But we have poor perception of yellows that intense, and such objects tend to be soft, lacking in detail. So it's something to keep in mind, yet not particularly serious.
In the book, I illustrated this with a picture of a yellow pepper whose brightest parts were quite close to 0c0m100y, and therefore out of the sRGB gamut. I printed this CMYK version side-by-side with one that had been converted to sRGB and then reconverted to CMYK. The numbers are considerably different but nobody in the pressroom could tell whether the two images were identical or not.
To show the *potential* problem, I swapped channels and produced one magenta and one cyan pepper. Once again, I converted the CMYK to and from sRGB and printed side-by-side. In both, particularly the cyan pepper, the damage, and the color shift, is clear. I commented that this illustrated the sorts of damage that would be incurred if any pictures actually existed with these kinds of magentas or cyans, but inasmuch as they don't (otherwise, I wouldn't have to manufacture a magenta and a cyan pepper), why saddle yourself with the disadvantages of a large-gamut RGB?
As list members know, I am no fan of either sRGB or Adobe RGB, but if those are the only two choices in the world, most CMYK-oriented users should adopt sRGB. The exceptions would be those who aren't good at color correction or don't have as much time as they would like to devote to correcting the images. These folks would likely get more vivid colors at the expense of detail, a fair tradeoff under the circumstances.
Certainly, there must be photographs somewhere that contain detail in areas that are out of the sRGB gamut but printable in CMYK. At some point in history, some tasteless designer must surely have created a garment in the same color as my magenta or cyan peppers. We should, however, be disinclined to make workflow decisions based on images that show up once every five years or so. If we find a CMYK-bound image for which sRGB isn't big enough, we always have LAB or Adobe RGB.
If would, of course, be senseless to use ProPhoto RGB when the normal output is CMYK. All it would do is add more nonprintable colors--Adobe RGB already covers everything that can be printed, and then some.
WHEN THE TARGET IS BIGGER.
Some of us are fortunate enough to have output conditions that have a greater gamut than that found in CMYK commercial presswork and conventional photo labs. Usually there are extra inks in play beyond the standard CMYK. In the early days it was believed that these extra inks should be vivid colors, like the orange and green of Pantone's Hexachrome system. Such inks have proven too clunky to control easily, and nowadays it's more common to find much subtler secondary inks, often pink and light cyan, sometimes others.
As even more inks are added the point of diminishing returns is reached. There is only so much that one can do with a subtractive process regardless of how many inks are in play, plus, as we all know, there are significant challenges associated with separating an RGB file into four channels, let alone eight or twelve. Unfortunately, it's a powerful sales tool: just as when one camera manufacturer comes up with a 10-megapixel camera somebody else feels they need to introduce a 50-megapixel models and to imply that any photographer who doesn't own one isn't a professional, when one printer manufacturer touts an 8-color printer somebody else comes out with one with 24 inks and says that anything else looks drab.
Vendor claims as to how much these extra inks add aren't completely baseless but they are greatly exaggerated. Most additional gamut comes from the quality of the paper, not the extra inks. Solid colors can be improved marginally; the real gain is in pastels, where light-colored inks can do wonders, provided that the vendor has a decent separation algorithm, which some don't. Particularly, the gamut charts that vendors provide are, in my experience, worthless because they don't represent what their own algorithms can achieve with real photographs.
If you frequently work with such a device, sRGB may be an inconveniently narrow working space, unless you're comfortable with LAB, in which case you could divert brilliantly colored images to that. Some of the heavily magenta images referred to above had sections that were out of the sRGB gamut, but they were also out of the CMYK gamut, so no harm, no foul. With a better inkset and good paper, if you shoot a lot of pink flowers and the like, that safety net may disappear.
Therefore, I think that ColorMatch RGB, or even something as wide as Adobe RGB, is appropriate for people who commonly have to worry about such output. Regrettably, in the last few years there has been advocacy of ultra-wide RGBs as working spaces. Such clumsy editing spaces create many handling problems. These have been discussed frequently on this list and there is no point in rehashing them. The question is whether there are ever any advantages that might compensate for having to put up with them.
This refers to editing only, not temporary storage. I frequently acquire images from Camera Raw into ProPhoto RGB, which is its resident ultra-wide space, because the resulting file is often the best for channel blending purposes, even if I'm headed for CMYK. But I make a copy of it and convert to a more rational RGB as the base file, and use the ProPhoto version only for blending.
We are speaking of photographs here. If you have some other plan, like creation of artificial images with colors far more intense than found in the original photograph, then all bets are off. Limiting it to photographs, I have said for a number of years that I have never seen a single one for which Adobe RGB is insufficient, regardless of output condition. In the second half of this post, we will look at two images that challenge this.
RGB Working Space, Part 2 of 2
Posted by: Dan Margulis
Sat Nov 11, 2006 10:35 am (PST)
This continues the discussion of the preceding post, which ended by questioning whether a natural photograph exists containing detailing in colors that can be printed on any existing output device, yet are beyond the gamut of Adobe RGB.
If one were ever to be found, it would, I wrote, almost have to be something yellow. No matter how many inks are in use, output devices still have trouble with reds, greens, and blues. There theoretically can be a problem in certain magentas and cyans, but few if any photographs exist that contain these colors. Only in yellows, which are much more common, is there a respectable chance that the original capture might contain something that can be printed yet Adobe RGB can't accommodate. But it would have to be not only a brilliantly yellow object, but one with heavy detail. We don't have good perception in yellows that intense.
We now turn to two images that are intended to suggest that an ultra-wide RGB is useful. Surprise! They're both strongly detailed yellow flowers. One is from Vladimir Yelisseev, the other from Andrew Rodney. I have extracted and posted several sub-images from each and posted them at https://www.ledet.com/margulis/2006HTM/RGB_Workspace_Images.zip
(BEWARE: It's a 60 mb file!) All variants are based on files opened with Camera Raw defaults into either sRGB, Adobe RGB, or ProPhotoRGB. Andrew has posted his raw images on his site. For space reasons Vladimir's raw image is not here, but it is on the CD of the new book [PP5E].
These files are large enough that anyone inclined to test them on printers should be able to do so. They are not JPEGged, in the interest of preventing artifacting.
DESCRIPTION AND WARNING.
Vladimir's image is of a flower with brick-reddish outer petals but an intensely yellow center, from which filaments emerge. Andrew's shows a yellow, but much paler, flower, also with prominent filaments. It is one of two substantially similar images that he posted in .DNG format. Because he was apparently unaware that .DNG files may derive from various sources, I requested that he repost the files in raw format. He did not post the same images, but two others that made it clear that the original .DNGs had not been altered before being saved. I chose this one as offering the fairest comparison to Vladimir's.
We have heard on this list how there is never any harm in acquiring images in ProPhoto RGB. The first set of examples demonstrates how wrong that idea is. In setting up this demonstration, when there's a question of which version is superior, the images are only identified by letter and are placed randomly. In this way, you are invited to decide which is better without being prejudiced by knowing in advance which is which. The identifications--if you need them--are kept in a separate text file, which is posted in the Files section of this group at http: //tech.groups.yahoo.com/group/colortheory/files
Nobody advocates working in sRGB if you have an extended-gamut printer; the only question is whether you should use something bigger (meaning, to most people, Adobe RGB) or whether an ultra-wide RGB is needed. It is, however, useful to *test* sRGB vs. ProPhoto, because any problems Adobe RGB might have would be worse in the sRGB version. If we conclude that sRGB is not sufficient, *then* we can look at whether Adobe RGB is good enough.
We'll start, oddly enough, in CMYK. Images A and B are Vladimir's flower, C and D Andrew's. In each case, one version was opened in ProPhoto and converted into CMYK using SWOP v2; the other opened in sRGB and treated similarly. It has been asserted that the two pairs should be identical. As you can see, they aren't. The images opened in sRGB are superior, particularly in Vladimir's shot. In Andrew's, you have to paste one image on top of the other and toggle back to see the difference, which is subtle (the ProPhoto version goes green in the shadows). Again, I'm not identifying which is which--you can get that information from the text file. If you agree that the sRGB-to-CMYK versions are better, this also suggests that printing the sRGB files to any printer without an extended gamut would also give better results.
What you see as composite on the monitor in these CMYK images is reasonably accurate. All remaining examples, however, are ProPhoto RGB files. Most are beyond the capability of your monitor to display accurately, so you need to look at the red and green channels to get an idea of what detail is lost. (We are insensitive to contrast variation in the blue channel, so that can be ignored.)
THE TEST RESULTS.
To know whether there is any point to using a wider gamut space requires looking at several factors. We need to know whether colors are actually present that the smaller RGB does not contain; whether any such colors can actually be printed by any known device; whether we would even wish to print those colors if possible, and whether the area contains significant detail. In the current images, the results are as follows:
1. Do any colors exist in the ProPhoto versions that cannot be replicated in the sRGB versions? Yes. Versions E and F show these areas, exaggerated to the same extent. Vladimir's image has a considerably larger area that encompasses almost the entire center of the flower and implicates the red channel even more than the blue. Andrew's is limited to the blue channel.
2. Are these colors achievable on any current printer? In Vladimir's case, probably, in Andrew's, unlikely. Versions G and H show how the CMYK versions vary from the ProPhoto originals. If we were to assume an extended-gamut printer, the shapes of the areas would be about the same, but they would be somewhat smaller.
Compare Versions E and G, and F and H. In Vladimir's case, there is only a fair correspondence between the two. Not all the areas shown as being out of the sRGB gamut are out of the CMYK gamut--and that's even assuming we are talking about offset presswork. If we assume something better, obviously this problem gets worse.
Andrew's image does not have this issue. The entire area that's out of gamut in sRGB is also OOG in CMYK, and quite a bit more. Even if it's area is significantly reduced there will be no problem.
This finding was confirmed by Murray DeJager, who tested it using the same inks and printer advocated by Andrew and Rich, and found no color difference between files output in sRGB and ProPhoto.
3. Do we really wish to achieve these colors? In Vladimir's case, some of them; in Andrew's, definitely not. Cameras don't see flowers the way we do. We break the colors apart (simultaneous contrast) and we dislike reproductions that overwhelm us with similar colors. Neither of these images is acceptable as shot for that reason--the both blast us with too much yellow.
In Vladimir's case, we want to tone down some of the center so that the brightest yellows seem more intense by comparison. I will talk about how to do that in a bit. The very yellowest parts, however, need to be as yellow as possible. So, if extra gamut is available, we want to make use of it.
In Andrew's case the flat yellow area is not an important part of the flower. It distracts from the pale yellow petals, which no printer has much of a chance of matching, but which need to be as yellow as we can make them.
4. Does the area contain significant detail? This is another way of asking whether opening in the narrower-gamut RGB can damage the file. Vladimir's image, yes; Andrew's, no.
When Andrew presented his image he suggested that running the Saturation slider of Hue/Saturation or Channel Mixer on the sRGB version of his file would show that it was inferior. These suggestions were so quickly debunked by other members of the list that little time need be spent on them. Channel Mixer is inappropriate for comparing *any* two RGBs because of the variance in channel structure. Running the Saturation slider on an areas that is already near the edge of a colorspace's gamut is something that most beginners would know not to do. Also, it is well known that increasing saturation works significantly better in LAB than any RGB, as verified by Murray's testing, where he reported that a saturation increase in LAB with Andrew's picture resulted in output clearly superior to either the sRGB or ProPhoto version.
If we wish to know whether opening a file in sRGB has damaged it in a way that ProPhoto would not, there is only one valid way of finding out, which versions J and K (Vladimir's) and L and M (Andrew's) allow you to test for yourself. In each pair, one of the unidentified variants was opened directly into ProPhoto. The other was opened into sRGB, but then converted in ProPhoto. You now need to ask whether there is anything you could possibly do to one variant that you couldn't duplicate in the other. In Andrew's case, the answer is no. For most purposes, the version acquired in sRGB and converted to ProPhoto is actually superior because the affected area has been desaturated slightly, which is desirable. There is no loss of detail except in the blue channel, which has no impact on overall contrast.
I conclude, then, that Andrew's image handles at least as well, and probably better, if acquired in sRGB as opposed to ProPhoto. There is therefore no point in testing it further. In Vladimir's, however, I conclude the opposite. All three channels are damaged in the sRGB version, especially the red. This could be repaired with some effort but obviously the scenario is not desirable. I therefore regard it as proven that at least one real-world image exists for which opening in sRGB causes problems if the destination is an extended gamut printer.
The next step, then, is to find out if the problem persists if Vladimir's image is opened in Adobe RGB rather than sRGB. Version N corresponds to the earlier Version E: it shows the area in which AdobeRGB version falls short of the one acquired in ProPhoto. Of course, the affected area is much smaller than the sRGB version, but it's still there, and it's still conceivable that the colors are printable, and we would still wish to achieve some of them if they are.
So the question boils down to what would happen next. I think that anybody who knows anything about color would attempt to add shape to the yellow area because as it stands there would just be a yellow blob in the center of the flower. The likely way to do this would be to lightly stamp the green channel into the luminosity of this area, while assuring that the very brightest parts become as yellow as possible. There are several ways to achieve this effect. If you would like to experiment with one, I provide version O, a two-layered blank document with the top layer at reduced opacity and with Blend If options enabled to restrict the move largely to the yellow areas.
In testing this, I used a ProPhoto file plus a ProPhoto to Adobe RGB back to ProPhoto file. The results were equivalent because the green channels of the two were basically identical. The approach would be to convert a copy of the image to LAB, increase saturation in the yellows to taste, and put copies on both layers of version O. Then, on the top layer, replace the L channel with the green channel of either of the RGB versions. Where to set opacity and Blend If is a matter of taste, I like the settings in Version O, but you should get the same results with whatever you like.
The last two paragraphs, however, are relevant to nothing, because a sensible person would not open that image into Adobe RGB. The moment I saw the preview of Vladimir's image in the dialog I knew that it might be that exceptional image for which Adobe RGB might not be sufficient. I did not know, as I do now, that I could get just as good a result from opening it in Adobe RGB. So, had it been a real job, I would not have wasted time finding out. I would simply have opened the file in ProPhoto RGB and gotten out of it when convenient.
Ultra-wide colorspaces serve an important role in conversions and file interchange. It's not dangerous to store a file in such a space if you are careful with what happens to it thereafter; Photoshop, for example, uses ProPhoto RGB as a reference space in Camera Raw. As editing spaces, they have grave disadvantages. This is as true for LAB as it is for ultra-wide RGBs. LAB, however, has many important advantages that make it suitable for many kinds of images. Ultra-wide RGBs have none. The argument, heavily subsidized by vendors whose profit depends on ink sales more than printers, is that the printers are so incredibly capable that only an ultra-wide RGB can feed them. If that were true, it might or might not be a justification for accepting all the disadvantages.
As these files show, however, it isn't true. Charts showing that certain printable colors are out of an RGB's gamut are meaningless if no pictures exist that attempt to access those colors. As I have previously pointed out, just about the only photograph that could require the use of an ultra-wide RGB and still potentially be printable would involve brilliant yellows containing critical detail. Vladimir's image meets that description, but it's a rare case. If you get a raw file of a brilliant yellow flower that has significant detail, you recognize it for what it is, and use ProPhoto to acquire it.
Meanwhile, you need to consider that the image being offered by the proponents of ultra-wide spaces is the best they can do--but it doesn't even need Adobe RGB, let alone something wider. Andrew's image works perfectly well in sRGB, possibly better than ProPhoto RGB.
If you are planning to force a file into the deep magentas and cyans that don't exist in photographs but are achievable on certain printers, you will get better quality from LAB, but an ultra-wide RGB may serve. However, those who think that they are getting something out of editing normal photographs in an ultra-wide RGB are only fooling themselves. If they are lucky, their quality won't be hurt too badly. If not, they'll wind up with clunky corrections and lots of problems with out of gamut colors. This is not a workflow for those serious about image quality, so I reiterate the traditional policy of this list that extended discussion of editing in exotic RGBs belongs elsewhere.