Luminous Landscape Forum
Raw & Post Processing, Printing => Colour Management => Topic started by: Rhossydd on December 02, 2009, 10:11:52 am
-
We frequently see comparisons of gamut, by volume or area, using various tools such as Color Think Pro, Gamutvision, Colorsync etc.
Can someone explain if there’s a standard deviation factor(eg + or - 1.2 DeltaE ) that defines where the gamut boundaries are ?
Is this enshrined in the ICC standards ?
Thanks
Paul
-
We frequently see comparisons of gamut, by volume or area, using various tools such as Color Think Pro, Gamutvision, Colorsync etc.
Can someone explain if there’s a standard deviation factor(eg + or - 1.2 DeltaE ) that defines where the gamut boundaries are ?
Is this enshrined in the ICC standards ?
Thanks
Paul
I don't understand the question. The boundaries of a theoretical gamut (sRGB, pro-photo, etc) are defined by the gamut itself. The boundaries of an empirical gamut (printer X, paper Y) are defined by measurement. Where would SD come into play?
-
I don't understand the question. The boundaries of a theoretical gamut (sRGB, pro-photo, etc) are defined by the gamut itself.
So at what measurement does the gamut deem to stop ?
How is the limit of gamut defined ? and what is the measurement tolerance ?
-
How is the limit of gamut defined ? and what is the measurement tolerance ?
I'll try to answer from what I've understood so far.. If someone knowledgeable can correct, please do.
For a peripheral, the gamut is defined as all the colors the device can output. As long as a color is produced it is in the gamut, the limit we see on most (all?) graphs is the set of the most extreme (the most saturated for a given luminosity, or vice-versa) of these colors.
The profiling target "just" has to parse the RGB values to make the device output all the possible colors.
Instrument precision does not take part of the definition of the gamut (of course, it still has an influence on the measured one).
-
For simplicity, let's consider only the two most common types of profiles: matrix and Look-up Table (LUT) based. A matrix profile is defined by a (comparatively) simple set of equations and parameters. Examples include Adobe RGB and ProPhoto RGB. The gamut boundaries of these profiles are smooth and easy to calculate. A profile of a real world device such as a printer uses a LUT. This is a table of points correlating a actual, real-world color to the corresponding color in the printer's color space. While matrix profiles are nicely behaved mathematical abstractions, LUT profiles are usually anything but.
Determining the gamut boundary of a real-world device takes measuring it. Errors occur due to measurement inaccuracies or, as is becoming increasingly common with wide gamut LCD displays and colorimeters such as the EyeOne Display, the range of the measurement device being insufficient to read the full range of the display. Any profile made under such conditions is guaranteed inaccurate, so worrying about gamut boundaries is the least of your problems.
Once you have a profile, computing the gamut boundary is a matter of number crunching. The gamut hull is computed at a finite number of slices. How these steps are performed, how coarse or fine the sampling grid, and the mesh size used to fit the gamut hull all factor in to how accurate the gamut calculation will be. A pathologically behaved printer requires a finer sampling grid to produce the same absolute accuracy of a gamut volume calculation than does a colro space such as PP RGB.
The other tradeoff is computation time. You can perform a very accurate calculation if you are willing to wait long enough. We use gamut hull calculations as part of our printer profiling software and went through many algorithm tuning iterations. We finally settled on a set of calculations that spits out a gamut volume in few seconds on our 8-core server boxes. This gives values within 1% of those produced by an hour long run. If there is interest I can cobble together a standalone program to calculate gamut volumes on a more typical hardware system with ~+/- 5% accuracy (Windows only -- sorry).
-
Maybe I'm not being clear enough with my question here. I'll try phrasing it differently;
How accurate are the gamut boundaries ? 2%, 5%, 10%...... ?
Presumably the factors involved are measuring device accuracy, device stability, rounding errors in profiling software calculations, plus is there a tolerance factor within the specification to allow for these inaccuracies.
-
Maybe I'm not being clear enough with my question here. I'll try phrasing it differently;
How accurate are the gamut boundaries ? 2%, 5%, 10%...... ?
Depends on the the accuracy of the measuring device. As to where the boundaries are located, that depends on the physical characteristics of the device (monitor, printer, etc.). With monitors, it depends on the CRT phosphors, LCD color filters and backlight source, or LED characteristics. With printers, it's a combination of the inks and paper.
-
Depends on the the accuracy of the measuring device.
Er, yes. I did say that already and added some other possible factors. I'm starting to wonder of anyone here actually knows the answer.
It seems an interesting issue to me when so much is concluded from gamut visualisations.
-
I'm starting to wonder of anyone here actually knows the answer.
For synthetic editing profiles (ProPhoto, sRGB, etc.), there is no measurement error--the gamut boundaries are an intrinsic part of the mathematical definition of the profile.
For measured profiles, you'd have to look up the accuracy specifications for the device that measured the profile data, if you know what device was used to make the measurements. If you don't even know the measuring device used, speculating about measurement accuracy is kinda pointless...
-
speculating about measurement accuracy is kinda pointless...
If you take that point of view or specification any review that references gamut data without qualification (generally not possible with manufacturer's own materials) is worthless.
This comes back to the original question, how accurate are profile gamut measurements and what tolerances are deemed acceptable.
So, when comparing gamuts how much can we conclude from them ?
For example, are transmissive measurements generally more or less accurate than reflective measurements. If there were a significant difference, comparing screen profile gamuts to other gamuts might be unreliable. Should we trust figures like "107% of AdobeRGB" when reading a monitor specification if the tolerances involved in building the profile are +/- 1.5% ??
-
Rhossydd,
How long is a piece of string?
Your question is very open ended, people here have tried to accommodate your original question. Perhaps if you become a little more specific and word your question differently, as in a way that include more information as to what you really want to know.
To answer the question " Should we trust figures like 107% AdobeRGB?" Do you have access to the procedure that the device was measured under, if so, and that this procedure satisfy you, then you have good reason to believe/trust that the device covers 107% AdobeRGB. Do you trust some who says their monitor covers 50% of aRGB? when you measure it crudely it does indeed cover 95% aRGB? <where crudely is with your own color device, under uncontrolled conditions>
I use a EIZO CG241W, NEC 2690 and a DELL 2407... I think the first two is rated by their respective companies to be at or around 95% or higher in terms of coverage of the aRGB color gamut. I will profile them with two trusty iOne's and the CG is just under, the NEC is over, by a degree more then the CG is under. The Dell however, is well under, but that I would also expect, as they have not made any attempt to tell me its close.
Trust.... what is your tolerance of trust? is it equal to, is it 0,00000000000000000000000000000......00000000001% or greater precision acceptable?
is 2-3% ok? is it binary, YES or NO? maybe 5-10% is acceptable?
Can you operate, at any of these tolerances when you work on any of these monitors, devices? can you tell the difference? Some can, maybe not with precision, but they can tell you that there is a difference. That difference may be to one, an advantage, a positive thing, but to another it is just not good enough.
So, in my own case, my latest monitor, an NEC 2690 has met and exceeded my expectations. I don't actually recall how big a gamut NEC claimed and I am too lazy to go looking now, but whatever the number was, it has given me enough confidence in the product to recommend it to you or anyone else who needs a nice Monitor, in its price range. Can it display 32bit color files NO. Did I expect it? NO Would i like one that could? YES! Can I afford one? NO So, the NEC fits the bill for me and I trust it to display enough or more of aRGB gamut for me to output or produce files for other people. Not that this has anything to do with NEC, but I have used it as an example here to illustrate my point.
I hope you find what you are looking for, and when you do I hope you are able to see it.
:-)
Henrik
-
This comes back to the original question, how accurate are profile gamut measurements and what tolerances are deemed acceptable.
So, when comparing gamuts how much can we conclude from them ?
For example, are transmissive measurements generally more or less accurate than reflective measurements. If there were a significant difference, comparing screen profile gamuts to other gamuts might be unreliable. Should we trust figures like "107% of AdobeRGB" when reading a monitor specification if the tolerances involved in building the profile are +/- 1.5% ??
You can get a rough idea of the accuracy of a profiling/measurement device by using it to profile your monitor and then visually comparing a Color Checker reference image file displayed on-screen to an actual Color Checker. The manufacturers of colorimeters and spectrophotometers do usually post measurement accuracy data, though it may take some digging to find it.
The other thing to consider is that converting DeltaE figures (which is what manufacturers usually quote, since that is the best indicator of how closely profiled colors match original colors) to percentages involves some pretty advanced math. Human vision is better at distinguishing between some subtle shades of color than others, and DeltaE takes this into account. If you want a specific percentage difference, you'd have to choose a particular shade of green, for example, and then choose similar, but slightly different green values and run them through the DeltaE formula and compare the percentage of change to DeltaE. If you do the same exercise for shades of orange, the results would NOT be the same, by design. And that's a good thing. Get over your fixation on percentages; percentages will not answer your questions in a meaningful way. A 2% change in blue is not the same as a 2% change in green is not the same as a 2% change in orange, etc, etc, etc.
Transmissive and reflective measurements are equally accurate if made with the same measuring device. There's no reason for there to be any difference. Why would there be?
The ultimate test of color management is this: Photograph something, say a garden gnome figurine. Display the photo of the gnome on your monitor, and print the photo with several different printers on a variety of papers. How closely do the colors of the original gnome figurine, the monitor image, and the various prints match? My experience with the EyeOne spectro is that you can get everything pretty darn close if you profile everything properly; close enough that a side-by-side comparison is necessary to distinguish subtle differences. DeltaE is a fairly good quantification of the magnitude of the differences you will see, but it has a non-constant relationship to the actual margin of error expressed as a percentage. My experience is that the manufacturer's claims of color accuracy (at least from GretagMacbeth/X-Rite) are trustworthy and translate well to real-world results.
The bottom line is that profile comparisons and other color management claims are generally reliable, trustworthy, and correlate reasonably well to real-world results.
-
You can get a rough idea of the accuracy of a profiling/measurement device by using it to profile your monitor and then visually comparing a Color Checker reference image file displayed on-screen to an actual Color Checker.
You're missing the point here. It's not about one system's accuracy it's about comparative accuracy.
The other thing to consider is that converting DeltaE figures........A 2% change in blue.... etc,
I was using the term percentage only as a general concept to try to help people understand what I'm trying to find out. Ultimately digital colour meters will only output numbers anyway, so a percentage error can be significant way of looking at it. How any percentage error is then handled when processed in software adds another potential tolerance factor.
The ultimate test of color management is this:... How closely do the colors of the original .., the monitor image, and the various prints match?.
No, not really. The real test is taking an image from an outside source and being confident on how it will be reproduced on another output device, eg on a magazine cover. This is what colour management was implemented to handle.
This comes back to how comparatively accurate profiles are.
Yes, I know that without knowing what device was used to create the profile there's an element of trust involved, but is there really no specified tolerances of accuracy overall within the ICC specification ? If not, how are we to trust this information at all ?
I'd rather hoped some of the experts here might have been able to shed some light on this, but maybe it's all taken on trust and not understood too well.
-
No, not really. The real test is taking an image from an outside source and being confident on how it will be reproduced on another output device, eg on a magazine cover. This is what colour management was implemented to handle.
Yes, really, at least for most of us on this forum. We are not involved in magazine production but rather in taking photos and wanting to create prints that match our vision. If the profiles I use let me do this (which they do), then academic discussions about "accuracy" become so much mental masturbation and a complete waste of time. If I was a carpenter and had a saw that let me cut wood in all the ways I wanted, I would hardly be concerned if the blade were actually 1/2 inch shorter than its design specs.
-
Yes, really, at least for most of us on this forum.
Statisticaly, I'm sure you're correct. Most people aren't interested in this subject, but this forum is one of the few web forums where one can ask difficult, possibly academic, questions about digital photography and stand some chance of getting an informed answer.
If you're not interested in the more esoteric discussions here, just ignore them, but don't try to 'dumb down' the forum just because you're not interested or don't understand the issues underlying a particular thread.
We are not involved in magazine production
No, but a greater proportion than most here DO need to consider the wider issues of colour management. Maybe not for magazines, as such, but consider the increasing number of photographers self publishing through companies like Blurb, Lulu, Snapfish etc. A look at Blurb's support forums suggests there's a lot confusion and misunderstanding about colour management when photographers starting to work with something more involved than their desktop printer.
-
No, not really. The real test is taking an image from an outside source and being confident on how it will be reproduced on another output device, eg on a magazine cover. This is what colour management was implemented to handle.
WTF do you think I'm talking about? If I can color-manage my workflow such that I can shoot a garden gnome figurine, and print images of the garden gnome that are essentially indistinguishable from the original on 5 different printers and 10 different paper stocks, then color management is doing its job and I can be confident that my colors will be correct when I print, regardless of whether it's a magazine cover or whatever, as long as I follow best practices for profiling, etc.
This comes back to how comparatively accurate profiles are.
Yes, I know that without knowing what device was used to create the profile there's an element of trust involved, but is there really no specified tolerances of accuracy overall within the ICC specification ? If not, how are we to trust this information at all ?
Let me repeat myself...Manufacturers DO specify accuracy tolerances, but in DeltaE instead of percentages.
I'd rather hoped some of the experts here might have been able to shed some light on this, but maybe it's all taken on trust and not understood too well.
It would help if you actually read and paid attention to what you've been told here, as most of your questions have already been answered...
-
as most of your questions have already been answered...
Not at all.
-
I think this thread got inexcusably nasty, just because the original poster didn't phrase his question clearly.
"We frequently see comparisons of gamut, by volume or area, using various tools such as Color Think Pro, Gamutvision, Colorsync etc.
Can someone explain if there’s a standard deviation factor(eg + or - 1.2 DeltaE ) that defines where the gamut boundaries are ?
Is this enshrined in the ICC standards ?
Thanks
Paul"
I think what Paul intended to ask was "Is there a specified tolerance for stating color gamuts somewhere in a standards document?" The implication is "Are stated gamuts of tight enough tolerance to be useful, and how picky should I be in comparing published gamuts?"
If these are the intended questions (Paul, please correct me if I'm wrong), the answers are:
1) there is no standard for how precise color gamut measurements should be
2) small differences in gamut on the order of a few Delta E are not significant anyway.
-
I think what Paul intended to ask was "Is there a specified tolerance for stating color gamuts somewhere in a standards document?" The implication is "Are stated gamuts of tight enough tolerance to be useful, and how picky should I be in comparing published gamuts?"
If these are the intended questions (Paul, please correct me if I'm wrong), the answers are:
1) there is no standard for how precise color gamut measurements should be
2) small differences in gamut on the order of a few Delta E are not significant anyway.
I answered both of those questions in my previous posts. If you want to know measurement accuracy, you have to know the measurement tool and its accuracy specs.
-
I think what Paul intended to ask was "Is there a specified tolerance for stating color gamuts somewhere in a standards document?" The implication is "Are stated gamuts of tight enough tolerance to be useful, and how picky should I be in comparing published gamuts?"
If these are the intended questions (Paul, please correct me if I'm wrong), the answers are:
1) there is no standard for how precise color gamut measurements should be
2) small differences in gamut on the order of a few Delta E are not significant anyway.
Thanks Wayne. A great first post here, welcome.
Yes, you've understood what I was getting at almost completely. My original question was loosely(poorly in hindsight) worded to also cover the possibility that different profile creation tools might be working to different tolerances if a specific tolerance isn't specified in the ICC standards.
1) Right thanks. I'll get round to actually reading these standards one day when my colour knowledge is better than it is now.
2) True, but manufacturers are promoting products on the basis of percentages of standard colourspaces. If errors in measurement are greater than the implied accuracy, those figures become rather futile.
The other issue is that gamut comparisons are often made via visualisation tools where DeltaE isn't used. I wonder if it's better to consider the gamut volume of devices more as soft edged 'clouds' rather than the tightly defined wireframes. If that makes sense, what would be useful to know is how fluffy the clouds are.
-
Rhossyd, ignoring instrument error and noise for the moment, in general the gamut limits (as shown in a gamut plot) are largely defined by the hardware setup and the number of samples measured. For example, when you see a gamut plot for a printer profile, what you're really seeing is a collection of measurements (at some sampling, e.g., 12 x 12 x 12 = 1728 samples, uniformly spaced in the device coordinate system) taken for a specific printer, paper, and ink, with a specific set of driver settings (and driver version). All of these have an impact on the results. The boundary in the gamut plot is the surface that contains the set of measurements ...
-
... in general the gamut limits (as shown in a gamut plot) are largely defined by the hardware setup and the number of samples measured.
The number of samples is surely more likely to effect the profile's accuracy, not the extent of gamut. Unless the sampling points have been very poorly chosen and don't test enough colours close to the gamut's limit.
For example, when you see a gamut plot for a printer profile, what you're really seeing is a collection of measurements (at some sampling, e.g., 12 x 12 x 12 = 1728 samples, uniformly spaced in the device coordinate system)
My understanding is that X-Rite (GMB) targets don't use a uniformly distributed range of sampling colours, but have a higher density of sampling points closer to the neutral axis. The uniform points are only within the profile, many of which may be created by interpolating the measurement data anyway.
The boundary in the gamut plot is the surface that contains the set of measurements ...
? Not sure that makes sense to me. Surely the majority of sampling points are within the gamut volume, not on the boundary.
-
The number of samples is surely more likely to effect the profile's accuracy, not the extent of gamut. Unless the sampling points have been very poorly chosen and don't test enough colours close to the gamut's limit.
Yes, but in general if you want to understand the shape of the gamut boundary, you need to have enough samples along the boundaries of the device coordinate space. For example, 8-bit samples along the boundary of a RGB device coordinate space might look like (0, 0, 0), (0, 0, 5), (0, 0, 10), ... (0, 0, 255).
What determines the gamut, as indicated earlier, is the printing system as a whole: printer, paper, ink, and driver settings. One thing I forgot to mention earlier is measurement geometry and assumed lighting conditions. Most colorimetry is based on 45/90 measurement and a D50 reference. If you view your prints under very non-D50-like light sources, the actual gamut would be very different.
My understanding is that X-Rite (GMB) targets don't use a uniformly distributed range of sampling colours, but have a higher density of sampling points closer to the neutral axis. The uniform points are only within the profile, many of which may be created by interpolating the measurement data anyway.
That depends on the target. For smaller targets, that is true. Some of the larger targets are uniformly spaced.
Not sure that makes sense to me. Surely the majority of sampling points are within the gamut volume, not on the boundary.
They are, but as indicated above, if you want an accurate plot of the boundary (e.g., when looking at a cross section at L* = 50), you need enough points on the boundary ...
-
What determines the gamut, as indicated earlier, is the printing system as a whole: printer, paper, ink, and driver settings. One thing I forgot to mention earlier is measurement geometry and assumed lighting conditions. Most colorimetry is based on 45/90 measurement and a D50 reference. If you view your prints under very non-D50-like light sources, the actual gamut would be very different.
My question is primarily about comparisons of profile gamuts and their boundary accuracy.
... if you want an accurate plot of the boundary ..... you need enough points on the boundary ...
This suggests that the number of samples used to build a profile and their range might be more important in defining gamut than any other issue.
This leads to other questions;
Is there a minimum number of samples needed to build an ICC profile ?
Is that number specified within the specification for ICC profiles ? If not, why not ?
How are we to trust the claimed gamuts promoted on manufacturer's data if the quality of the profiling isn't declared ?
-
Er, yes. I did say that already and added some other possible factors. I'm starting to wonder of anyone here actually knows the answer.
It seems an interesting issue to me when so much is concluded from gamut visualisations.
Actually I think you may be overcomplicating the concept.
If I'm profiling a printer, I'm going to send a series of colors and print them ... many of those are going to exceed the gamut of the printer. So whatever color results is the maximum that printer is capable of. I can't get a redder red or a bluer blue, etc
I measure that result so I now have an actual number of the devices maximum capability, and using that data can create a profile to defined that boundary and map all of the colors into the printers space.
How accurate? sure there are things that can affect that accuracy, and some devices are better than others. So you might see a profile built with a colorMunki as having a slightly smaller gamut volume than one built with an i1 Pro, but all the devices are very capable.
In the end, you have a gamut volume defined by testing the printer's capability, and your color management system is going to remap the image colors into that color space, so everything that gets sent to the printer is within it's capabilities. The overall gamut isn't a guess, but empirically measured to a highly accurate degree.
In fact I would be extremely surprised if most people can tell the difference between a print made with a profile from a colorMunki and one made using better devices. While perhaps the better device more accurately measures the colors, those measurements are at a level far beyond our ability to see them. (I just did this test, and out of about 20 people so far, they results are split ... and everyone of them admitted to just guessing because they really couldn't see a difference). the color munki profile is built by sampling only 100 colors, the i1 profile 1728.
-
My question is primarily about comparisons of profile gamuts and their boundary accuracy.
Understood. Just keep in mind that most plots are done with a D50 reference illuminant, so the "accuracy" of such a plot may not be meaningful to you. For example, comparisons between two gamut plots under D50 does not tell you how the gamuts of those two printing systems compare under a different lighting (e.g., an office fluorescent bulb under which prints may actually be hung).
This suggests that the number of samples used to build a profile and their range might be more important in defining gamut than any other issue.
No, in practice there are sufficient samples in most profiling systems to get a reasonable description of the gamut boundary. The primary factors that actually determine the gamut in practice are the main variables in the printing system, as indicated earlier: printer, paper, ink, and driver.
This leads to other questions;
Is there a minimum number of samples needed to build an ICC profile ?
No. Technically, you can build an ICC profile without any samples.
Is that number specified within the specification for ICC profiles ? If not, why not ?
The ICC specification itself is not concerned with the quality of a profile. The specification is concerned with the format of the profile (i.e., how the description of a device can be interpreted in a portable, platform-independent way) and workflows. It is up to the creator (designer) of a profile to provide details about the quality of a profile and the methods used to build it, if he or she wishes.
-
No, in practice there are sufficient samples in most profiling systems to get a reasonable description of the gamut boundary. The primary factors that actually determine the gamut in practice are the main variables in the printing system, as indicated earlier: printer, paper, ink, and driver.
You're still assuming I'm talking just about printer profiles, but most of my concerns are with comparing printer profiles with monitor profiles.
There seem to be potential differences between the accuracy of each type of profile, eg are reflective measurements more or less accurate than emisive measurements ? Screen profiling systems usually use far fewer samples for measurement than printer profiles, so are they likely to be less accurate ?
Technically, you can build an ICC profile without any samples
Which of course would be useless for characterising a device's output in the real world.
-
You're still assuming I'm talking just about printer profiles, but most of my concerns are with comparing printer profiles with monitor profiles.
There seem to be potential differences between the accuracy of each type of profile, eg are reflective measurements more or less accurate than emisive measurements ? Screen profiling systems usually use far fewer samples for measurement than printer profiles, so are they likely to be less accurate ?
Which of course would be useless for characterising a device's output in the real world.
Printer profiles need more samples than monitor profiles because they are typically more non-linear than monitors.
THe number of patches used to create the profile is based on the number needed to create a good profile. A lot of really smart people have put a lot of effort into figuring this stuff out, and most CM stuff works as advertised if you know how to use it properly. You're overanalyzing this stuff.
-
2) True, but manufacturers are promoting products on the basis of percentages of standard colourspaces. If errors in measurement are greater than the implied accuracy, those figures become rather futile.
The other issue is that gamut comparisons are often made via visualisation tools where DeltaE isn't used. I wonder if it's better to consider the gamut volume of devices more as soft edged 'clouds' rather than the tightly defined wireframes. If that makes sense, what would be useful to know is how fluffy the clouds are.
Trying to make it clearer: the statements for volume of gamut are plenty accurate enough, since you are not going to notice the difference unless it is large (like a 10% or more increase in maximum saturation). The only real problem with these comparisons is that they are often done as a percentage of the NTSC television standard area on a CIE 1931 chart, which doesn't do the best job of telling you how much perceivable volume you have lost or gained, or whether it's in important areas of the chart. So don't worry about how fluffy the clouds are.
On the other hand, you need to be concerned with precision if you are trying to determine if the primary colors of a display exactly match sRGB (or whatever standard you are after). Mismatching the primaries does make a difference in the displayed colors.
Another thing that gets left out of these discussions is the variability of color vision from viewer to viewer. The most accurate possible display only matches for the "standard observer." People with normal color discrimination can vary in the extreme by as much as +/- 20% in the ratio of red and green primaries to match a given yellow wavelength - and the wider the gamut of the display, the larger the variation. So, chasing the last decimal place is a hobby or a research project. A truly accurate color reproduction system would need a camera tuned to your personal cone sensitivities (color matching functions) running a display visually matched by you to your desired white point.
-
We frequently see comparisons of gamut, by volume or area, using various tools such as Color Think Pro, Gamutvision, Colorsync etc.
Can someone explain if there’s a standard deviation factor(eg + or - 1.2 DeltaE ) that defines where the gamut boundaries are ?
Is this enshrined in the ICC standards ?
Thanks
Paul
I'll take a shot at answering this. So regarding the accuracy of the 3d gamut projections, I can say that there is a bit of wiggle room for displaying gamut boundaries. First there is really no standardized way to do this. A simple method for drawing the boundary is to round trip color data e.g (lab > rgb > lab) and then plot the most saturated colors at different luminance steps. Now depending on the rendering intent used and the quality of the profile the accuracy of this projected gamut can certainly vary. There is also going to be a bit of a tolerance between the ICC profile's numerical representation of the printer's gamut, and what the actual printed result will be. Conversion "noise" when round tripping in gamut color can easily be about 2∆e, and you can expect another few ∆e between the expected results and the actual results. So the 3d boundary may be a fairly precise map of the ICC gamut but should be considered an approximation of the printer's gamut. The ICC profile is essentially a translation table and it's good at answering questions in the format of Given A what is B. For example trying to do something simple like determining if a single color is out of gamut is not an exact process.
So yes, when you are comparing two profiles, especially a printer and monitor profiles you can expect a bit of wiggle room regarding the actual boundaries of the profiles. Trying to come up with a general percentage as to how much this varies would be difficult since the linearity of the printer, quality and precision of the profile are going to vary widely. It's also going to vary by location in the color gamut. Also comparing gamut volumes is a poor tool for comparing two profiles as the difference in volume might be all in one area of could be spread throughout the profile and two different software packages could create profiles with two different gamut volumes from the same set of measurements.
To respond to a few other things:
There is not a direct correlation between sample patches and profile accuracy or precision. When building a profile most advanced software packages give you the ability to specify the internal bit depth of the profile and the size of the grid used in the LUT. You could measure a target with 3K patches and make a profile with fewer grid points, than a profile made with 50 patches. Also there is not a direct correlation between the measurement patches and the points in the LUT.
Typically printer profiles are LUT based, and monitor profiles are Matrix based.
The ICC spec is built around the D50 illuminate and 2˚ observer.
There is no standard regarding the number of patches needed for building a profile. Most software gives you an option. With RGB profiles I've found diminished return after 729 patches on most modern printers. When building CMYK profiles I generally use the 1617 patch IT87.4 target.
Determining the outer boundaries of a printer with reasonable accuracy does not take that many patches. If a device is fairly linear we can easily draw a line between two colors to extrapolate the gamut of the printer(which is what all 3d plots are doing.)
For very non linear devices more patches can be better. For a very linear device more patches can actually cause problems, especially if it is used for photographic purposes. HP when building the profiling software for the Z series printers was able to use very few patches and get fairly good results because they hardwired certain characteristics of the printer into the profiling model, having only to account for different media types. With general profiling packages they have to be able to account for radically different devices from offset presses and laser printers to inkjet and dyesub printers.
There is little correlation between profile accuracy and measurement device accuracy. A decent device like the i1 are calibrated to a standard well under 1∆e. Using the same software I would not expect measurements from a Munki vs an i1 to produce a profile with any important differences. When trying to certify proofs, or measure color standards where your tolerances are under 1∆e then instrumental differences such as measurement interval (10 vs 20 nm) or light source (xenon vs led) can be important.
-
Thank you Julian for that well explained reply.
Also comparing gamut volumes is a poor tool for comparing two profiles the difference in volume might be all in one area of could be spread throughout the profile
I'm afraid there's room for some confusion here. You can either 'compare volumes' by the crude tool of percentages, as monitor manufacturers are prone to do, but you can also 'compare volumes' visually with a visualisation tool such as Color Think or Gamutvision which should reveal far more useful information with respect to gamut limitations between devices.
two different software packages could create profiles with two different gamut volumes from the same set of measurements.
That would surely imply that one package would be failing to deliver correct results, yes ?
-
Thank you Julian for that well explained reply.
I'm afraid there's room for some confusion here. You can either 'compare volumes' by the crude tool of percentages, as monitor manufacturers are prone to do, but you can also 'compare volumes' visually with a visualisation tool such as Color Think or Gamutvision which should reveal far more useful information with respect to gamut limitations between devices.
That would surely imply that one package would be failing to deliver correct results, yes ?
Agreed, generating a gamut volume "by the numbers" is certainly a valid metric, although one that has a margin of error, and a degree of interpretation. For example if you want to find out the gamut volume of the sRGB color space, depending the tool you use, you may see the volume listed as 821K (∆e3) to 909K or some other number, and this is for a simple synthetic RGB profile. The volume is somewhat open to interpretation, as far as I know there is no standardized way of generating gamut volume nor is this something encoded in the ICC profile. So just looking at numbers doesn't tell you much. Visualizations of gamut are a much better tool, but treat them as a model rather than as a perfectly exact representation of the actual gamut of a device.
Regarding the second question, two packages can(and do) produce different results, and that is the expected behavior. The ICC spec only says, "hey, this is how you build and ICC profile" It sets how information should be encoded and organized, and what has to be included. How those LUT's are generated are up to the profiling software. Lets say I measure eight patches consisting of primaries, overprints a black and a media sample, (white, black, red,green, blue,cyan, and magenta,yellow) From that info I could build a profile. Of course to I would have to interpolate a lot of information but I could get something very crude. Obviously there is a lot of room to use different algorithms to fill in the blanks, and as long as I build a profile according to the ICC spec all would be valid. Of course this gets exponentially more complex when you have to make decisions about how to generate a "neutral" gray on a very blue paper, or how to create the perceptual rendering intent table. There is no strict procedural process for creating the table in a profile so some very smart people have come up with very novel and different ways of creating profiles, this secret sauce is why we own copies of Profiler and ProfileMaker. The two packages produce different flavors of profiles, the same way that a camera loaded with two different types of film (hmm is this analogy going to be unusable in another decade) or two different digital cameras will render a scene slightly differently, and yet both are equally correct(or incorrect).