Gamut confusion

[gkroeger]

Having thought I understood soft proofing, gamuts and rendering intents, I now discover my ignorance.

I have an image which, when soft proofed in PS (CC 2017.1.1) show no highlighted pixels (even at 100%) when Gamut Warning is enabled. I am soft proofing with a profile from Hahnemüle Photo Gloss Baryta for the Epson SC-P9000.

My understanding was that Rendering Intent controlled the way out-of-gamut colors were mapped into the printer gamut and how in-gamut colors were moved by that mapping. As such, it would seem to me that if the image has no out-of-gamut colors, there should be no difference between Relative Colorimetric and Perceptual appear. That is not the case, as changing the Rendering Intent changes the appearance of the image on my calibrated NEC PA302W monitor.

Adding to my confusion, when I plot this image along with the profile in 3D with ColorThink 2.0, it shows out-of-gamut colors in the image.

What am I missing?

Thanks,
Glenn

[Mark D Segal]

When you are viewing the photo on display with the different Rendering Intents, how are you setting Black Point Compensation and Simulation of Paper White?

[howardm]

People 'in the know' know that PS's OOG display is reasonably inaccurate (like needing a 6dE error) and is a very 'blunt' tool.

I know digitaldog has a video on this topic.

[gkroeger]

When you are viewing the photo on display with the different Rendering Intents, how are you setting Black Point Compensation and Simulation of Paper White?

Mark:

Normally, I have both checked... but I have tried all permutations and still get changes in appearance when I change rendering intent.

Glenn

[gkroeger]

People 'in the know' know that PS's OOG display is reasonably inaccurate (like needing a 6dE error) and is a very 'blunt' tool.

I know digitaldog has a video on this topic.

I suspect the "blunt tool" explanation is correct. Interestingly, if I choose Hahnemüle's profile for the Canon Pro-2000 I get lots of out-of-gamut pixels, but when I choose the profile for the Epson SP 9800, which has a smaller gamut, I get no out-of-gamut pixels.

[Mark D Segal]

Reply #2 is correct in the sense that the tool is a blunt way to show OOG colours, because you can't use it to see the extent to which a colour is OOG. It's binary: in/out. However, I'm not sure this is the issue here. I suspect it is more related to the profile. Recall, a softproof is generated by a reverse flow of information through the tables in the profile. It could be (but not necessarily) that the manner in which these tables are performing isn't entirely consistent between Rendering Intents. That said, if you review Adobe's definitions of what these Rendering Intents do, it would appear that the two Intents can still produce slightly different looking outcomes even if all the colours are in gamut.

[gkroeger]

Reply #2 is correct in the sense that the tool is a blunt way to show OOG colours, because you can't use it to see the extent to which a colour is OOG. It's binary: in/out. However, I'm not sure this is the issue here. I suspect it is more related to the profile. Recall, a softproof is generated by a reverse flow of information through the tables in the profile. It could be (but not necessarily) that the manner in which these tables are performing isn't entirely consistent between Rendering Intents. That said, if you review Adobe's definitions of what these Rendering Intents do, it would appear that the two Intents can still produce slightly different looking outcomes even if all the colours are in gamut.

Thanks Mark... you are correct... upon rereading Adobe's description of rendering intents, there can be different mappings even for in-gamut colors. As for the gamut warning, like you, I suspect that there is something different about some of Hahnemüle's profiles since the difference between profiles in visual appearance is minuscule but the difference in gamut warning is enormous.

[bjanes]

As such, it would seem to me that if the image has no out-of-gamut colors, there should be no difference between Relative Colorimetric and Perceptual appear. That is not the case, as changing the Rendering Intent changes the appearance of the image on my calibrated NEC PA302W monitor.

Current CMMs are not smart with respect to handling gamut compression in perceptual intent rendering in that they apply compression without looking at the actual gamut of the image. The amount of compression depends on the profile lookup tables, but there will be some compression of the more saturated colors even if none of them are out of gamut. See Steves Digicams for a nice discussion, which is quoted in part below:


Other than reduced color vibrancy, there is another down side to perceptual intent. The current ICC CMM (color management model) is not a "smart" model meaning that it cannot and does not examine the gamut of the actual image before trying to compress it to fit in the printer's gamut

Regards,

Bill

[Doug Gray]

Current CMMs are not smart with respect to handling gamut compression in perceptual intent rendering in that they apply compression without looking at the actual gamut of the image. The amount of compression depends on the profile lookup tables, but there will be some compression of the more saturated colors even if none of them are out of gamut. See Steves Digicams for a nice discussion, which is quoted in part below:


Other than reduced color vibrancy, there is another down side to perceptual intent. The current ICC CMM (color management model) is not a "smart" model meaning that it cannot and does not examine the gamut of the actual image before trying to compress it to fit in the printer's gamut

Regards,

Bill

Yes, this is correct. In fact, i1Profiler can create innumerable, different Perceptual Intents based on a couple slider settings. There is no specification as to how a gamut is mapped in Perceptual. It is both different and there are significant variations between the Perceptual intent construction with different vendors. It's something one adjusts to taste or just leaves at default.  OTOH, Relative Intent is well defined for all colors that are in gamut. Options in I1Profiler that impact Relative are limited to a few that impact accuracy such as the number of LUT cubes, interpolation smoothness, and 8 or 16 bit tables. The largest factor affecting profile accuracy is the number of patches used in creating the profile together with the printer's native driver's characteristics.

Photoshop's OOG masking is useful but only with Relative Intent and will show image parts that exceed 6 dE. Perceptual Intent doesn't have a specific metric for OOG. Even though Photoshop's OOG mask will still show up on more extreme colors it will not show areas where significant compression exists and just looking at the soft proof will be more productive with Perceptual.

[gkroeger]

Thanks Bill and Doug... these insights help!

[Wayne Fox]

My understanding was that Rendering Intent controlled the way out-of-gamut colors were mapped into the printer gamut and how in-gamut colors were moved by that mapping. As such, it would seem to me that if the image has no out-of-gamut colors, there should be no difference between Relative Colorimetric and Perceptual appear. That is not the case, as changing the Rendering Intent changes the appearance of the image on my calibrated NEC PA302W monitor.

While logical, I think you might be making one assumption that isn't true.  Just because all of the colors are within gamut doesn't mean there won't be anything done to the colors.  There is not a lot of logic built into profiling that determines "if" you need to push colors into gamut.  So the colors will get moved based on the rendering intent whether or not the image contains colors that are out of the devices gamut.  Thus a perceptual intent will almost always affect colors within gamut, even if your image has no colors out of gamut.  I tested this years ago, by printing an image of various colors, and then eliminating the out of gamut colors and reprinting.  There wasn't a perceptual change in the colors whether or not out of gamut colors were included.

so this is more of a tool to help select which rendering intent to use.  If indeed you have no colors out of gamut, then using relative is probably the more accurate intent since it will pretty much leave the colors alone.  If you have a lot of colors out of gamut, then indeed perhaps perceptual may be a better option.

[Schewe]

While logical, I think you might be making one assumption that isn't true.  Just because all of the colors are within gamut doesn't mean there won't be anything done to the colors.  There is not a lot of logic built into profiling that determines "if" you need to push colors into gamut.  So the colors will get moved based on the rendering intent whether or not the image contains colors that are out of the devices gamut.  Thus a perceptual intent will almost always affect colors within gamut, even if your image has no colors out of gamut.

Wayne is correct...even if all colors are in gamut, RelCol and Percept will not match.

[pfigen]

My inkjet workflow *used* to be, back in the day, to convert to profile and then print with no color management, which let me, when I needed to, convert two copies of the file with different rendering intents and then blend them together for printing. That doesn't seem to work anymore but it was a great method of having perceptual rending just in the areas that needed it without altering the rest of the image. I still use that quite often for cmyk press work and it's remarkably effective. Hell, you could even combine relative from one vendor with perceptual from another that way. I remember sharing that with Joseph Holmes on the phone many years ago and he asked if he could "borrow" that technique. Of course.

[Doug Gray]

My inkjet workflow *used* to be, back in the day, to convert to profile and then print with no color management, which let me, when I needed to, convert two copies of the file with different rendering intents and then blend them together for printing. That doesn't seem to work anymore but it was a great method of having perceptual rending just in the areas that needed it without altering the rest of the image. I still use that quite often for cmyk press work and it's remarkably effective. Hell, you could even combine relative from one vendor with perceptual from another that way. I remember sharing that with Joseph Holmes on the phone many years ago and he asked if he could "borrow" that technique. Of course.

You still can.
http://forum.luminous-landscape.com/index.php?topic=117516.msg974007#msg974007

[Ethan Hansen]

Wayne is correct...even if all colors are in gamut, RelCol and Percept will not match.

True. How Perceptual rendering functions is up to the profiling software. In general, the goals are to make a visually pleasing conversion between color spaces. This implies minimizing banding as colors clump near the gamut boundary and, for some implementations, smoothing tonal transitions, opening shadows, etc. All this color manipulation means all colors are altered to some extent, otherwise you will get banding in other areas.

Rel. Color prints in-gamut colors unchanged. What happens at gamut boundaries again depends on the profiling code. Some implementations collapse similar colors into one, while others take more liberty with the rendering intent and smooth transitions even if it is not mathematically the most accurate approach.

Something older versions of Photoshop were particularly poor at was displaying the effect of contrast range compression. The IQ 180 back I used today has a dynamic range of ~13.5 giving a contrast ratio of more than 12000:1. Good ink jets printing on glossy stock reach 300-400:1 while matte paper often has a print contrast under 100:1. Traditional wet lab printing as still used by Fuji Frontier and similar printers max out around 100:1 as well. Obviously something has to give.

Photoshop's simulation options attempt to show the effects of contrast range  The gamut warning often does not reflect what is actually happening.

A further issue is what source color space is assumed perceptual algorithm. As far as I can tell, i1Profiler uses either sRGB or a slightly expanded gamut version thereof. Current ICC specs attempt to address this problem, but that has been the case for years. Don't hold your breath.

Images with significant amounts of out of gamut colors, particularly ones in larger color spaces, benefit from device link profiles. Unlike normal profile conversions a device link can be created using the actual image gamut for the source color space rather than the much larger mathematical abstraction of ProPhoto, etc. If you are comfortable going the command line route, Argyll is a powerful option.