sRGB and aRGB - Printable and non-Printable Colors

[Doug Gray]

Here I compare gamuts of Canon Matte and Baryta Fine Art Natural (a semigloss) on a Canon 9500 II.

The purpose was to determine colors that are in sRGB or aRGB (Adobe RGB (1998)) but are not printable. In addition colors that are printable but are outside sRGB and examined. 8 histograms of dE2k are attached in groups of 4, one for each paper type, for those that are curious about the dE2k distributions.

I started with all colors in ProPhoto, 8 bit RGB space then removed colors that are outside the ICC LAB PCS spec. of a and b limited to -128 to 127. Then the subset of those colors that exist within the colorspaces sRGB and aRGB were selected. The printer gamut was examined by roundtripping using Colorimetric intent and selecting dE2k of < 2 as being within gamut. For sRGB and aRGB inside gamut status was assigned if the dE2k was < 0.1 since they are matrix based and have clear boundaries.

Here's the results for the sRGB and aRGB colorspaces

Percent of Legal ICCLAB colors in ProPhoto RGB: 87
About 13% of the ProPhoto RGB values produced values that get clipped by the ICC profile algorithms. Some of these are real colors, and some are "imaginary" due to ProPhoto extending beyond the human gamut in places. Some "imaginary" colors are also legal ICC Lab values. Since illegal Lab values get clipped they appear in tests to be the same color. They were removed as duplicates.

Percent of sRGB in ProPhoto and icclab: 25.1
25.1% of the colors exist within sRGB colorspace. A profiled monitor that has primaries that are at or beyond the sRGB primaries can display all of these colors to accuracies of typically better than 2 dE2k depending on the monitor and software.

Percent of aRGB in ProPhoto and icclab: 36.7
36.7% of the colors exist within aRGB colorspace. A profiled, wide gamut monitor that has primaries that are at or beyond the aRGB primaries can display all of these colors to accuracies of typically better than 2 dE2k depending on the monitor and software. Most wide gamut monitors come close and can accurately display 95% or more of these.


Here's the results for Canon Matte Paper.

Percent sRGB colors not printable on the 9500 II:  37.1
This is the percentage of the sRGB colorspace that cannot be printed to better than dE2k of 2. This is fairly high, even though we are only using sRGB because matte paper has a small gamut.

Percent aRGB colors not printable on the 9500 II:  48.5
Almost half of aRGB colors cannot be printed on this matte paper.

Percent of extra colors printable on the 9500 II relative to sRGB:   7.2
Relative to the entire sRGB working space, there are 7.2% more colors that could be printed but aren't when using sRGB because they are outside the sRGB gamut.

Percent of extra colors printable on the 9500 II relative to aRGB:   0.1
Only a tiny fraction of colors that can be printed on matte are outside the aRGB space.


Now we turn to Baryta, a wide gamut, non OB, semigloss paper. We can see that significantly more of both sRGB and aRGB colorspaces can be printed. Interestingly, there is about the same percentage of sRGB colors that can't be printed as colors that are outside of sRGB but are printable. Also, aRGB expands the gamut a lot but there are still 4.6% more printable colors that aRGB can't handle.


Here's the results for Baryta high gamut paper.

Percent sRGB colors not printable on the 9500 II:  22.1
Percent aRGB colors not printable on the 9500 II:  33.3
Percent of extra colors printable on the 9500 II relative to sRGB:  22.0
Percent of extra colors printable on the 9500 II relative to aRGB:   4.6

[Doug Gray]

Here's a set of patches that show a few effects on my 9500 II using the Baryta paper. They are all in 8 bit ProPhoto RGB though the first two sets are within sRGB and aRGB respectively.

The attached a set of patches that show typical colors that:

1. Are within sRGB but not printable (they are outside the printer's gamut)
2. Are within aRGB but not printable (they are outside the printer's gamut)
3. Are printable but outside sRGB's gamut
4. Are printable but outside aRGB's gamut.

The patches were selected based on the following criteria. The are arranged in order of decreasing L values.
a) Each patch in a group must have dE2k > 10 from all other patches in the same group, and
b) The dE2k must be >= 5 between what is requested and what is printed
 
1/6/16 -Chart replaced. The labels for 2 and 3 were swapped.
1/6/16 -Printable Colors outside of sRGB replaced to make dE2k the same as other patches

[Bart_van_der_Wolf]

Here I compare gamuts of Canon Matte and Baryta Fine Art Natural (a semigloss) on a Canon 9500 II.

The purpose was to determine colors that are in sRGB or aRGB (Adobe RGB (1998)) but are not printable. In addition colors that are printable but are outside sRGB and examined. 8 histograms of dE2k are attached in groups of 4, one for each paper type, for those that are curious about the dE2k distributions.

Hi Doug,

Thanks for sharing. This reconfirms that a perceptual rendering intent, and soft proofing, can be really useful. Also, not all colors in actual images cover the full aRGB or sRGB colorspace.

Most real image colors have a modest saturation, and will probably map just fine to actually printable colors. Some printing applications also add (optional) dithering, which complicates the analysis, but covers up lots of perceptual issues such as posterization. And with proofing, we may be able to reduce the issues that do exist.

Cheers,
Bart

[Doug Gray]

Hi Doug,

Thanks for sharing. This reconfirms that a perceptual rendering intent, and soft proofing, can be really useful. Also, not all colors in actual images cover the full aRGB or sRGB colorspace.

Most real image colors have a modest saturation, and will probably map just fine to actually printable colors. Some printing applications also add (optional) dithering, which complicates the analysis, but covers up lots of perceptual issues such as posterization. And with proofing, we may be able to reduce the issues that do exist.

Cheers,
Bart

Hi Bart,

Indeed, not that many real life colors are outside aRGB or even sRGB. Mostly I see the gamut pushed with flowers, some fabrics and brightly colored consumer product labels.

Perceptual intent typically does often provide the most attractive prints but suffers from lack of consistency from one profile vendor to another. Even within a vendor, you can typically tweak Perceptual.  What those tweaks do, exactly, is poorly documented if at all. So soft proofing gives a way to see exactly what you will get. Same with Relative and the other Intents. With Relative you should expect to be able to produce visually identical prints no matter what profiler software is used as long as the colors are in gamut. If the colors are outside, soft proofing lets you see what those colors are turned into.

Generally, for precision work, I prefer Relative Intent. So long as the colors are in gamut they will be reasonably accurately printed. It provides more consistency between different profiles and media since Relative is supposed to be measurement based and most profile makers adhere to that.

[bjanes]

Here I compare gamuts of Canon Matte and Baryta Fine Art Natural (a semigloss) on a Canon 9500 II.

The purpose was to determine colors that are in sRGB or aRGB (Adobe RGB (1998)) but are not printable. In addition colors that are printable but are outside sRGB and examined. 8 histograms of dE2k are attached in groups of 4, one for each paper type, for those that are curious about the dE2k distributions.

Here's the results for Baryta high gamut paper.

Percent sRGB colors not printable on the 9500 II:  22.1
Percent aRGB colors not printable on the 9500 II:  33.3
Percent of extra colors printable on the 9500 II relative to sRGB:  22.0
Percent of extra colors printable on the 9500 II relative to aRGB:   4.6

Doug,

Thanks for posting these data. The extent to which the relatively small color space sRGB contains colors that are not printable with wide gamut ink jet printers is larger than many realize and your results here are quite informative and indicate the need for soft proofing.

I'm not certain how you derived your data, but another way to look at these gamuts is with 3D plots using Colorthink or similar software. Shown below is a plot of the sRGB gamut compared to that of the Epson 3880 with Premium Glossy photo paper. The Epson supplied profile was used. The printer gamut is considerably larger in the mid-luminance yellows and greens, but the sRGB gamut is larger at high luminances. IMHO, these high luminance colors contribute considerably to the impact of the print, probably more so than the expanded gamut of the printer in the mid-luminance yellows and greens. Soft proofing with an sRGB monitor will show the appearance of the high luminance colors in the print and the inability to print these accounts for what Jeff Schewe calls "make my image look like crap" when the simulate paper color is checked is checked with Photoshop softproofing.



The image below shows gamuts for the Epson and NEC PA241W monitor, whose gamut is close to Adobe RGB.  There is considerable gain in the mid-luminance gamut, but many mid-luminance yellows and greens that are in gamut for the printer are out of gamut for the monitor and can not be successfully soft proofed. I don't have quantitative gamut data, but the out of gamut AdobeRGB/NEC colors that are printable appears somewhat larger than your data indicate.



I agree with your finding that many flower colors are well outside of Adobe RGB gamut. Shown below is a tulip photographed with the Nikon D800e and rendered into ProPhotoRGB using ACR with the Adobe Standard Profile and no saturation boost other than that provided by the standard profile. It is uploaded in ProPhoto and will not look good when viewed with a non-color managed browser. The saturation is hardly unusual.



The colors are well outside of Adobe RGB as shown below:



Regards,

Bill

[digitaldog]

I'm not certain how you derived your data, but another way to look at these gamuts is with 3D plots using Colorthink or similar software.

That's the ideal way to do these tests! How CTP plots gamuts is also important.
http://www.colorwiki.com/wiki/Color_Management_Myths_26-28#Myth_26
Doug doesn't apparently own CTP, a shame, it's really the tool to use here.

[Doug Gray]

I'm not certain how you derived your data, but another way to look at these gamuts is with 3D plots using Colorthink or similar software.

I used Matlab with the image toolbox. I have some functions that let me easily explore profile characteristics. For instance:
abs_lab_image=ProfileConvert(ref_image, 'EUPPG 9500 II TC9.18 with 25Grays RGB.icm', 'f', 1, 'r', 3);
Will render an image in Lab using relative colorimetric intent (the 'f', 1) then convert it back to the actual colors printed, (the 'r', 3);
Matlab makes it easy to quantify color changes using a variety of scenarios.

The colors are well outside of Adobe RGB as shown below:

Bill, I downloaded the 4226_ProPhoto.jpg but it is tagged as sRGB. If I assign ProPhoto it looks really awful on my wide gamut monitor.  Am I correct in ignoring the sRGB tag and assigning ProPhoto? When I do that I have to desaturate 30% or so for the colors to start to look reasonable.

For instance if I examine the pixel at X=46, Y=93 I get RGB values of 213,0,36 which is Lab (52.8,125.1, 56.8 ). This particular Pixel has the largest deltaE2000 of 14.1 distant from Adobe RGB (1998) of any pixel in the image.

[Tim Lookingbill]

Since Bill mentions setting the camera profile of the tulip image to Adobe Standard in ACR, has anyone considered the gamut model of that camera profile?

The reason I say this is for my camera Adobe Standard makes for a great gamut taming (saturation throttle) on such subjects as vibrant flowers and sunsets compared to a custom DNG profile. I would think seeing that much of a color change in a test image as Bill presented should consider how it influences gamut controlled saturation throttling appearance models.

[Doug Gray]

That's the ideal way to do these tests! How CTP plots gamuts is also important.
http://www.colorwiki.com/wiki/Color_Management_Myths_26-28#Myth_26
Doug doesn't apparently own CTP, a shame, it's really the tool to use here.

I actually do own CTP. Bought it 4 years ago. It's useful for many things, particularly things that are baked in. It also has some nice export capabilities and beautiful graphics. However, for examining detailed statistical spreads or generating information on profile accuracy, it is somewhat limited. Most of my interest is in printable color accuracy in RI and AI. Also understanding how Perceptual Intent is mapped via compression and expansion near gamut boundaries. For instance, critical to me is how accurate the A2B1 tables are. CTP has some use in comparing relative accuracy between different vendor profiles of the same device and paper but doesn't tell you much about the specific accuracy of a given profile. Also, CTP will let you run device RGB values and tell you what the profile reported Lab color is but you need to actually print device RGB patches and measure them then compare that to the profile reported values. Patchtool (or Matlab scripts that do a similar job) are outstanding at this job. The error histograms that I've posted are from this process.

An example of the power of Matlab is the short program that identified colors that are out of the printable gamut but within Adobe RGB or sRGB.  It went through the entire constellation of possible colors, Selects the highest dE2k color, then selects the next highest dE2k that is at least 10 dE2k away and repeats that ensuring all selected colors are at least 10 dE2k away from each other to create a patch set. This is a reasonable representation of the actual colors that are out of gamut instead of listing perhaps a million or so individual, close colors. It can go through the entire ProPhoto RGB set of 16M, 8 bit colors in a couple minutes.

[bjanes]

I used Matlab with the image toolbox. I have some functions that let me easily explore profile characteristics. For instance:
abs_lab_image=ProfileConvert(ref_image, 'EUPPG 9500 II TC9.18 with 25Grays RGB.icm', 'f', 1, 'r', 3);
Will render an image in Lab using relative colorimetric intent (the 'f', 1) then convert it back to the actual colors printed, (the 'r', 3);
Matlab makes it easy to quantify color changes using a variety of scenarios.

Bill, I downloaded the 4226_ProPhoto.jpg but it is tagged as sRGB. If I assign ProPhoto it looks really awful on my wide gamut monitor.  Am I correct in ignoring the sRGB tag and assigning ProPhoto? When I do that I have to desaturate 30% or so for the colors to start to look reasonable.

For instance if I examine the pixel at X=46, Y=93 I get RGB values of 213,0,36 which is Lab (52.8,125.1, 56.8 ). This particular Pixel has the largest deltaE2000 of 14.1 distant from Adobe RGB (1998) of any pixel in the image.

Doug,

Thanks for the information. The matlab approach is very interesting, but I am not among the elite that have and are proficient in that program.

Regarding the image, the hosting software apparently converts the downloaded file to sRGB and resizes it.

Here is a link to the file on Adobe Cloud
http://adobe.ly/1mFRTIS

Regards,

Bill

[bjanes]

Since Bill mentions setting the camera profile of the tulip image to Adobe Standard in ACR, has anyone considered the gamut model of that camera profile?

The reason I say this is for my camera Adobe Standard makes for a great gamut taming (saturation throttle) on such subjects as vibrant flowers and sunsets compared to a custom DNG profile. I would think seeing that much of a color change in a test image as Bill presented should consider how it influences gamut controlled saturation throttling appearance models.

Tim,

For my D800e, Adobe Standard seems to apply a slight saturation boost as shown by Imatest.



Regards,

Bill

[Jim Kasson]

Here I compare gamuts of Canon Matte and Baryta Fine Art Natural (a semigloss) on a Canon 9500 II.

The purpose was to determine colors that are in sRGB or aRGB (Adobe RGB (1998)) but are not printable. In addition colors that are printable but are outside sRGB and examined. 8 histograms of dE2k are attached in groups of 4, one for each paper type, for those that are curious about the dE2k distributions.

I started with all colors in ProPhoto, 8 bit RGB space then removed colors that are outside the ICC LAB PCS spec. of a and b limited to -128 to 127. Then the subset of those colors that exist within the colorspaces sRGB and aRGB were selected. The printer gamut was examined by roundtripping using Colorimetric intent and selecting dE2k of < 2 as being within gamut. For sRGB and aRGB inside gamut status was assigned if the dE2k was < 0.1 since they are matrix based and have clear boundaries.

Good work, Doug. One thing that would probably make the color counting more meaningful would be to start with a regular 3D grid of colors in a perceptually uniform color space like Lab or Luv.

You could use your technique and create an image with all of the colors in 8-bit Lab, and go on from there.

That way the color count will relate to the volume in Lab.

If you want to ship me the code and have me do the work, that's fine, too.

Thanks for doing this.

Jim

[Doug Gray]

Good work, Doug. One thing that would probably make the color counting more meaningful would be to start with a regular 3D grid of colors in a perceptually uniform color space like Lab or Luv.

You could use your technique and create an image with all of the colors in 8-bit Lab, and go on from there.

That way the color count will relate to the volume in Lab.

If you want to ship me the code and have me do the work, that's fine, too.

Thanks for doing this.

Jim

Good idea Jim. I used the ProPhoto since so many people operate in that space and Adobe uses it internally but using Lab space could be more meaningful. Certainly the volume numbers would be a much better match with what's typically used for looking at gamuts.

It's close to trivial for me to change the code to run that but I appreciate the offer. I assume you have Matlab and the Image Toolbox.

[Doug Gray]

Regarding the image, the hosting software apparently converts the downloaded file to sRGB and resizes it.

Here is a link to the file on Adobe Cloud
http://adobe.ly/1mFRTIS

Regards,

Bill

I ran your image to pull out colors that were outside Adobe RGB. Here's a set of all the patches that are at least 5 dE2k different from each other and are at least 3 dE2k away from the Adobe RGB gamut edge.

[Jim Kasson]

Good idea Jim. I used the ProPhoto since so many people operate in that space and Adobe uses it internally but using Lab space could be more meaningful. Certainly the volume numbers would be a much better match with what's typically used for looking at gamuts.

It's close to trivial for me to change the code to run that but I appreciate the offer. I assume you have Matlab and the Image Toolbox.

OK, have at it. I'm interested.

Yes, In case we collaborate in the future, I have Matlab and the Image, Statistics, and both Optimization toolboxes. I also have optprop.

http://www.mathworks.com/matlabcentral/fileexchange/13788-optprop-a-color-properties-toolbox

If you don't have that last one, I recommend it. It's free.

Jim

[Doug Gray]

OK, have at it. I'm interested.

Yes, In case we collaborate in the future, I have Matlab and the Image, Statistics, and both Optimization toolboxes. I also have optprop.

http://www.mathworks.com/matlabcentral/fileexchange/13788-optprop-a-color-properties-toolbox

If you don't have that last one, I recommend it. It's free.

Jim

You have more than me. I have the image and standard optimization. Most of my stuff is signal processing and control  theory related.

Here's the results running Lab from L=0:100, and b from -128:127

Percent of icclab that is included in sRGB: 12.8
Percent of icclab that is included in aRGB: 18.5

Lab is a much larger gamut but many of the colors are not legal though some colors extend beyond ICCLAB which clips at -128/127


Percent sRGB colors not printable on the 9500 II:  20.2
Percent aRGB colors not printable on the 9500 II:  29.8
Percent of extra colors printable on the 9500 II relative to sRGB:  26.0
Percent of extra colors printable on the 9500 II relative to aRGB:   5.3

The percentages change a bit from using ProPhoto RGB but the ratios are relatively similar.

I've got the out of gamut patches too (very similar) but I have to paste them together and I have to run to a dinner and play.

Thanks for the optprop link. I did not have that. Sounds interesting.

[Jim Kasson]

You have more than me. I have the image and standard optimization. Most of my stuff is signal processing and control  theory related.

Here's the results running Lab from L=0:100, and b from -128:127

Percent of icclab that is included in sRGB: 12.8
Percent of icclab that is included in aRGB: 18.5

Lab is a much larger gamut but many of the colors are not legal though some colors extend beyond ICCLAB which clips at -128/127

It would be nice to prune the non-colors from the Lab image. The only way I can think of to do that is to calculate the xy values for each lab value (lab2xy if you're using optprop), then somehow figuring out if each xy value is inside the horseshoe. What am I missing here?

Jim

[Doug Gray]

There are two major areas of Lab pruning. The first would be, as you suggest, limiting Lab to the human xy surface. This would encompass all legal emissive device colors but with the caveat that there exists legal emissive colors outside of Lab.  The second, would be the MacAdam limits which are the theoretical limits of a reflective surface. Lab does encompass these. For MacAdam limits the most saturated colors, except for yellow, occur at the lowest luminosities. Gamut comparisons of printers are best analyzed in this space since it is the boundary which no printer can ever exceed. I've considered coding the relatively simple algorithm in this paper:

http://webserver2.tecgraf.puc-rio.br/~mgattass/fcg/Color/aic_1MacAdamLimits.pdf

Interesting to note that sRGB contains colors, as Bruce Lindbloom has shown, that are outside the MacAdam boundary. So it isn't surprising that there are many unprintable sRGB colors in real printers.

[Doug Gray]

Here's the detailed results from running all 8 bit Lab colors. The histograms are the counts in 8 bit Lab "cubes" for each of the 4 categories. Unprintable colors in sRGB and aRGB and colors that are printable but outside sRGB and aRGB. The X-axis is the deltaE 2000 which is an estimate of the perceptual distance between the actual color and the achievable color.

These colors are based on Relative Intent (so L=100,0,0 is scaled to paper white point and is within gamut). Printer was Canon 9500 II with Baryta Fine Art Natural which has a pretty high luminance white point but no OBs.

The attached image in 16 bit ProPhoto are the values of all colors that at least 10 dE2k different from each other and at least 5 dE2k different from their associated boundary. Note that 16 bit ProPhoto does not exactly represent 8 bit Lab values which can be off by +/- .01 or so. If you examine them in Photoshop using 32 bit mode Lab, values are shown to 2 decimals of precision.

[xpatUSA]

Bill, I downloaded the 4226_ProPhoto.jpg but it is tagged as sRGB.

But, as such, it is a very good example of gamut boundary-clipping (or whatever this forum calls it):



Check out the right-hand side of the saturation histogram . . .

Show Image is quite a quick tool for examining that kind of of stuff (only does JPEGs though).