There's nothing to fix—what you are seeing is correct.
When working in RGB, the three separate R,G,B channels work together to generate the final composite image. The green plate contributes about 60%, the red plate about 30% and the blue plate about 10%. In a neutral gray RGB image (completely desaturated), the red, green and blue channels must be identical, (if you are working in a normal gray-balanced workspace), otherwise you would have a color cast. When R=G=B, you have a neutral gray by definition.
If you click on the red, green and blue channels individually, you will see they are identical to one another. When you click on the composite RGB image it will darken, since the composite is built up from all three underlying channels. If you are using RGB, you can get a 2.2 gamma by using sRGB, Adobe RGB and some other working spaces. Gray gamma 2.2 will give you the same appearance, but it will use only a single channel. In a grayscale working space, your grayscale channel IS the image so you won't see this difference.
It is generally a bad idea to set your working space to Monitor RGB (at least if you plan to share you work with anyone else). Monitor RGB describes YOUR monitor, period. If you are doing mostly neutral RGB images for the internet, I'd select sRGB (which is the internet and email standard and has a 2.2 gamma tone curve). sRGB, having three channels, allows you to add some color, tone your images, etc. If you don't need or want that capability, and want to guarantee your images remain totally gray, then select Gamma 2.2 grayscale (single channel).
Why do you even care what the individual RGB channels look like?
Doesn't the gray working space determine how grayscale channels split to RGB when pasting as RGB pixels? As in assigning a profile never hurts until you convert?
Gray 2.2 as grayscale working space, single channel pasted as RGB pixels:
Dot gain 20% as grayscale space, single channel pasted as RGB pixels:
I have my gray working space set to 2.2 just for resolving this issue. If there a way to replicate the 2.2 behavior and not have to jump through hoops when working in actual single color images?
I'm not sure I understand your question/statement or what you are trying to achieve.
I just ran a little experiment (tell me if I am on the right track or whether I am completely missing your point, which is quite possible). I took an sRGB color image. I duplicated it so I had two identical sRGB images. One I converted to dot gain 20% using Relative Colorimetric with BPC on. The other was converted to Gray Gamma 2.2 with RC & BPC. Of course, the numbers changed, since a dot gain space is not the same as a gamma curve space, but the appearance is the same.
Then, I "selected all" and copied the 20% dot gain image and pasted it into a new sRGB document. I repeated the procedure for the gray gamma 2.2 image. This gave me two new sRGB images that are grayscale in appearance (fully desaturated).
All four images (whether dot gain 20%, gray gamma 2.2, or the two desaturated sRGB images look identical to me when switching back and forth from one to another. You can't "assign" from one mode to another (for example you can't assign a grayscale space to an RGB document). If you assign a different working space to your grayscale or RGB image, the numbers remain the same, but the appearance changes.
I have my "Color Settings" in Photoshop set to 20% dot gain, because I don't do a lot of grayscale internet work. Press work is more up my alley so I prefer to use dot gain. I also have all my color management policies set to Preserve Embedded Profiles. Maybe changing color settings will have some impact on the above scenario (I haven't tested that in ages, so I don't remember). I almost always convert (Edit > Convert to Profile) by selecting the exact profile and settings I want anyway instead of just using the "Mode" feature. That way, I know exactly what is going on and have control over the final outcome.
If I was doing a lot of B&W work for the internet, I'd probably set my gray working space in color settings to gray gamma 2.2, and use either sRGB or Adobe RGB, since they are both 2.2 gamma working spaces.
I'm rattling on, hoping I am understanding your question or statement.I'm not "jumping through any hoops" and all 4 of the above images have identical appearance. I must be misunderstanding something, or perhaps out color settings are different.
I'm talking about pasting a grayscale channel (any one will do, but obviously if it has some midtones it will help illustrate the problem) from a working doc into an RGB layer in the same doc. If your working space is not the same gamma as your RGB doc, the actual pixel values of the file (the RGB "grayscale" layer) are different than if you do have Gamma2.2 as your grayscale working space. Which seems to me to be problem.
Like for instance pasting the green channel as a luminosity layer ala Margulis, would result in two different luminosities, depending on your grayscale working space. Even if the whole time you worked in an RGB doc (never changing it's color space). I would argue the luminosity blend is correct if you gray space is 2.2, and incorrect if it is not.
The "hoops" I'm talking about amount to converting all grayscale images (to gain x) before printing. No biggie.
I have a bit clearer picture of what you are saying.
I just tried an experiment. In Color Settings, my RGB space is set to sRGB and my gray space is set to dot gain 20%. I created a new sRGB document, did a black to white gradient, and posterized it to give me 11 steps of gray from pure black to pure white.
I then created a new blank layer, ran Apply Image on the blank layer (using the background layer, green channel as a source, Normal mode, 100%). This pastes the green channel into the composite RGB channel of the top layer. Clicking that layer on and off results in no changes at all in the numbers, even though my default grayscale space is 20% dot gain. Using the Red or Blue channels gives me the exact same result.
Then, I tried another experiment. Leaving the above image on screen, I then went into color settings, trying different grayscale settings. The composite image does not change in appearance when going from dg30% to dg10% to gamma 1.8 or gamma 2.2. However, the underlying channels get lighter or darker with the grayscale setting used in Color Settings. I suspect this is what you are talking about. I don't find ANY standard grayscale setting that makes the individual channels match the appearance of the composite image. Some are closer than others, but none is a match. I suspect you'd need a custom dot gain curve for that. I'm not even sure why someone would want that, but I am sure there are probably some valid scenarios.
It was always my understanding that a composite RGB image was approximately 6 parts green, 3 parts red and 1 part blue from the standpoint of establishing luminosity. If you curve the blue channel in luminosity mode, you can go wild without altering the luminosity very much, but if you apply the same curve to the green channel, you get a huge luminosity shift. Red is in the middle.
I find all this interesting and like knowing what is going on 'under the hood', but I am not sure what it is we are trying to solve?
I think this is a problem with pasting only, not apply image.
Right you are! Pasting takes the appearance of the individual channel you copy (based on your current grayscale setting) and pastes it into the composite. Gamma 2.2 comes reasonably close to the composite image, but isn't the same (particularly in the shadows). If you could load L* as your gray working space in Color Settings, instead of gamma, I suspect it would be right on. Still, gamma 2.2 is probably the best overall choice for gamma.I think the only way to get closer would be a custom dot gain curve.
Thanks for pointing that out.
Paste is exactly the same if your color policy is preserve embedded profiles (for grays). I have mine to preserve and ask when pasting set (which of course within the same doc doesn't "work"/ask).
If you have them set to convert, then the shadows issue comes up.
I find this problematic if you save grayscale versions of your file as channels, but I don't think this is a huge issue. I was more worried that I am working around the problem by creating a new one (a la monitor space as working). I think as long as I convert my grayscales destined for press to the appropriate gain I'm OK, but I haven't sent anything to a color-managed printer to check. And all work seems to move away from single-color printing anyway.
I'm sure this is the OP's issue, but my solution could easily be problematic.
Martin Evening recommends it in his big CS4 for Photog book:
Hmmm....that's not what I am seeing. I have all my Color Settings set to
preserve embedded profiles, and pasting the green channel into a new layer
(in the same document) does not give the same appearance or the same
numbers. It's most noticeable in the shadow end of the tone scale. In my
11-step neutral wedge (sRGB), 25 becomes 18, 76 —› 75, 178 —› 179, 229 —›
230. So the dark end of the scale is getting noticeably darker, and the
light end is a tad lighter. Looks a lot like a curve plot comparing Lab to
2.2 gamma to me (as on Bruce Lindbloom's site, using his companding
Apply Image results in zero change in the numbers or appearance.
Doesn't affect me, and I don't mean to beat a dead horse, but so far, my
data doesn't agree. Maybe our settings are different?
I think you're right. I'm not sure what I was seeing as I can't replicate it. So maybe more with the apply image for me next time... Thanks for the help Lou.
I'm not sure who helped who? But I learned a few things and enjoyed the dialog.
Interestingly, I duplicated my original step wedge onto a new layer, then I went to Adjustments > Channel Mixer. I set it to monochrome, then set the red and blue channels to 0, and the green channel to 100%. Same result as Apply Image—no change. Just paste does funky things.
Talk with you soon, I hope, Lou.
Gamma 2.2 comes reasonably close to the composite image, but isn't the same (particularly in the shadows).
I think you need to use AdobeRGB1998 for your experiment since sRGB doesn't have a transfer function that is exactly gamma 2.2. See the graph at the end of this web page for a visualization.
Thanks Larry, that's perfect.
Larry, you're absolutely correct. Adobe RGB shows no difference between individual channels and the composite when viewing totally desaturated (B&W) images.
I didn't expect that, since both sRGB and Adobe RGB are designated as 2.2 gamma spaces. I created the same 11-step grayscale wedge in Adobe RGB, then converted to sRGB and I saw the numbers change to reflect my earlier post. And of course, assigning sRGB changes the display....mostly in the shadow areas. Interesting. So, Adobe RGB is "truly" 2.2 gamma, and sRGB is fairly close to 2.2 gamma.
Who'd have thunk it! Thanks.
BTW, Wide Gamut RGB (another true gamma 2.2 space) works perfectly, just like Adobe RGB. sRGB is odd man out!
I don't seem to have the wide gamut RGB on my computer (got deleted acidentally?) so I wonder if you could send me me a copy of Wide Gamut RGB as a color profile file? email address is firstname.lastname@example.org
Actually there IS a problem with the Adobe RGB (maybe wide gamut RGB would do better).The Adobe RGB doesn't accuratly convert certain low gray values correctly. What I mean is this:
When Gray is 255
R is 255
G is 255
B is 255
When Gray is 123
R is 123
G is 123
B is 123
When Gray is 24
R is 24
G is 24
B is 24
When Gray is 7
R is 7
G is 7
B is 7
BUT When Gray is 12
R is 13
G is 13
B is 13
and When Gray is 13
R is 14
G is 14
B is 14
Adobe RGB color profile ALWAYS messes up for Gray at 12 and 13.
This small artifact might not be a problem for a photographer who sees photography a just a form of art, but my interest is in image processing for scientific analysis (I have astrophotography for a hobby). So for me, bad numbers = corrupted data. I should be able to take a grayscale image and go FREELY between RGB and Gray modes with NO CORRUPTION of data.
On a related note, is there a way to make copy and paste copy the VALUES instead of the appearence? I thought this stuff was all taken care of when I turned OFF color management. I guess not. So my best bet might be to use Wide Gamut RGB.
The conversions from RGB to gray gamma 2.2 must introduce some interpolation error. Maybe the actual points of the gamma curves aren't identical between spaces. If you set your eye dropper to point sample, and blow up to a high magnification, even a solid block of a single density gray will show variations of 1 or 2 levels (after conversion, that is). For example, a block of solid gray (14R/14G/14B) in Adobe RGB is totally made up of level 14 pixels in the originating document, but averages out to 13/13/13 when converted to gray gamma 2.2. Running your eye dropper over the converted block shows many 13 and 14 level pixels, but using a 3x3 or 5x5 average sample size in the eye dropper, shows up as 13. There must be more 13 pixels than there are 14. So, perhaps it is the decimal precision used in Photoshop conversion engine when passing through the profile connection space (xyz or Lab). The Apple engine gives the same results, by the way. You'd think two different 2.2 gamma spaces would not result in any changes in these grays, but it does, even in 16 bit. Even the Apply Image command (which I used to transfer one of the Adobe RGB channels to the gray gamma 2.2 document) showed the same results.
If you need a grayscale for your scientific purposes, can you simply use one of the three RGB channels in the original RGB document? I haven't found a way to get precise numbers during conversion from RGB to grayscale. No big deal for my photographic purposes, because my eyes could never detect the difference, but it might be a big deal for scientific purposes.
I am curious to know what is actually happening behind the scenes. I'm not sure WG RGB is the answer. Converting from Adobe RGB to WG RGB game me similar interpolation errors and mixed pixels.
Lou & Benhut,
For some reason I'm unable to replicate. Can you provide simple idiot-proof
step by step instructions?
Hi Larry. Here is what I did to understand and simulate Benhut's problem. First, my color settings in PS are ARGB, gray gamma 2.2, color management set to preserve embedded profiles, all warnings checked, etc.
1. New Adobe RGB document. I created 5 or 6 blocks of dark neutral colors, separated by a little white space. I started with 10/10/10 and went up to 16/16/16 in 1 step increments. Just so I could see what was going on clearly, my original document was approximately 80 pixels wide by 10 pixels high so I could zoom in on individual pixels and see what was happening.
2. I tried a bunch of different ways to create a pure grayscale document in gray gamma 2.2 space.
a. Mode grayscale (with my grayscale space in color settings at 2.2 gamma)
b. Edit > Convert to profile > gray gamma 2.2, RC, BPC (tried both on and off). I tried all the other intents too.
c. Opened a new gray gamma 2.2 document and pasted both composite and individual channels into the grayscale document.
d. Opened a new gray gamma 2.2 document, and used Apply Image using the original RGB doc as source, Normal, 100%
I tried a few other things as well, but can't remember them all.
I could not get the destination document to leave the pixels values alone. I think it was level 14 that got clobbered the worst, and ended up average 13 in the new document. The others retained the "average" of the original, but you can see that not all pixels have the same level, even in a single block. They always seem to be interpolated, even though both spaces are 2.2 gamma. I even tried turning off color management in Color Settings, but still got the same result. So, I am guessing we have precision and interpolation error passing through the profile connection space. I even tried it in 16 bit and the numbers change from source to destination. After doing a 16 bit conversion, I changed to 8-bit, but ended up with the same basic problem. I could see it clearly when blowing my image up to a large magnification and running across the solid blocks with eye dropper set to point sample.
I'm using CS4 on a Mac, OSX 10.6.3. I don't need this level of precision and have no problem with slight rounding, but for scientific purposes, perhaps it does matter. But, I am curious.
Thanks Lou. I got level 14 in aRGB to become level 13 in Gray Gamma 2.2 after converting. In fact, I got level 14 to become level 13 in multiple ways:
-- aRGB to Gray Gamma 2.2
-- Gray Gamma 2.2 to aRGB
-- Gray Gamma 2.2 to Gamma 2.2 RGB (made up in Photoshop)
-- Gamma 2.2 RGB to Gray Gamma 2.2
This may be the result of slope limiting imposed by ACE (see last page of this pdf).
I've not tested for this possibility.
(A note of thanks goes to Gernot H for reminding me of slope limiting via private mail.)
Yup...same result. Thanks for the PDF. I figured it probably had something to do with interpolation, single or double precision decimals, the conversion engine, shape of the TRCs, or something like that. The same thing happens with the Apple CMM.
Good thing I don't have to rely on perfect transposition of data for my work (as picky as I am!!).
You talk about random differences, but that shouldn't be happenening, because I turned OFF the dithering for the color space conversion. And my gradient I drew is using the gradient tool with it set to linear (0% smoothing) and with the dithering turned OFF, and on a 256x256 canvas so there's exactly 256 levels in the gradient. This way I can check EVERY level of gray during conversion from Gray to RGB (or alternatively copying an R, G, or B channel and pasting into the RGB combined image). In spite of this, I still get the error I was talking about. I can NOT find out how to get the Copy+Paste function to AVOID copying the appearence, and instead copy the numbers. So far the only way to get a gray image perfectly into an RGB color space is with the "apply image" command.
I had photoshop on another computer too and didn't have this problem. Copy paste (and Gray to RGB) worked perfectly and copied the appearance just as here from Gray to RGB (or a single color chanel to RGB) but ALSO copied the numbers PERFECTLY! I don't remember what color space I was using, but I think it was 2.2-gamma for Gray and Wide-Gamut-RGB for RGB.
Anyone have a copy of color profile file of Wide-Gamut-RGB they could send me? For some reason it isn't on my current installation of PS or I accidentally deleted it somehow. My email address is email@example.com
I don't think Wide Gamut RGB is the answer since level 14 still converts to
level 13, but I'll send it to you anyway (under separate email).
If the Adobe Color Engine imposes slope limiting that is too stringent,
perhaps you can change the color engine to Microsoft ICM. Converting Gray
Gamma 2.2 level 14 to aRGB, for example, yields level 14 -- so it looks
promising. Perhaps you can report back here to let us know what happens to
all 256 values when you use this color engine?
I tried the Wide Gamut RGB and it didn't solve it. I tried the alternative to the Adobe conversion method, and it was even MORE inaccurate in the conversion. I tried the different conversion algorithms Relative Colormetric, Absolute Colormetric, and Perceptual with no fixing the problem. I tried differnt combinations of these different settings but with no fixing of the problem.
Thanks, Benhut. In that case there is probably no way to get what you need
via copy and paste. I am not aware of any way to turn off color management
totally in Photoshop. My suggestions for further exploration:
a) Ask over at the Photoshop forum;
b) Build a Photoshop action around the function that works, e.g., Apply
c) See if Filtermeister is a possibility (i.e., write your own filter so
that you can do even more).
Larry....some interesting findings.
I don't find that Apply Image gives accurate transposition of numbers when moving from Adobe RGB to gray gamma 2.2. Looking at levels 12 through 16 (up close, using point samples), I find that level 14 simply disappears, no matter what method I try. It becomes either level 13 or 15, but no 14.
So, I created a new gray gamma 2.2 document. I set my forground and background colors to 12/12/12 and 16/16/16 (using RGB mode in the color picker). Then, I drew a foreground to background gradient and inspect the pixels. My info palette was set to display grayscale values on the left, and RGB values in the right window. This gradient was created directly in the gg2.2 document, not brought in from the Adobe RGB document. Looking at the pixels up close with a point sample eyedropper, again, level 14 vanishes completely. It is not present in any individual pixel.
This could simply be rounding and/or the way PS presents the data, because grayscale data is presented as 0-100%, whereas rgb is presented as 0-255. Or it could be precision used in calculating the data, or the curves themselves. But, it goes beyond that. I displayed the grayscale info in 16 bit and it fluctuates quite a bit, in a fairly inconsistent manner. Maybe it is dithering, smoothing, curves, or the slope, as you mentioned. So, it doesn't seem the precision is there, at least for Benhut's scientific purposes.
For photographic work, it is more than accurate enough, in my opinion. But it is a curious phenomenon. I know you are a pixel-peeper like me, so I thought you might find this interesting.
It is possible. Remember when I said that on my desktop computer (the first copy of Photoshop installed), I'd managed to make it work setting it up properly somehow. But then more recently I installed it on my laptop, but I don't remember exactly what I did to make it work on my desktop. So best bet will be to go to my desktop and check what I had set it at. Won't be able to do that till I get home, because now I'm at college using my laptop.
Won't be able to do that till I get home, because now I'm at college usingmy laptop.
Okay, let us know what you find. The current level of precision more than
good enough for my purposes, but you now got me interested.
I don't find that Apply Image gives accurate transposition of numbers whenmoving from Adobe RGB to gray gamma 2.2
I'm not able to replicate your problem with levels 12 through 16. Here's
what I did:
a) Created a new aRGB document
b) Set up the info palette to show Actual Color (should avoid color
conversion in the Info Palette, I think...)
c) Switched the precision to 32-bit so I didn't have to do the math when
checking the grayscale values (levels 0..255 becomes 0..1)
d) Added solid patches with levels 12..16
e) Created a new document of the same size, gray gamma 2.2
f) Ran Applied Image using the aRGB image as source
g) Readouts are same as source, e.g., 0.047, 0.051, 0.055, 0.059, and 0.063
for levels 12 through 16.
Can you replicate my result?
I had to reboot my computer. I had an external drive that was doing weird things, and this can sometimes make OSX become weird or unstable, so I don't know if that skewed any of my previous results or not.
I duplicated your experiment, and I concur that Apply Image gives identical data when using the 32 bit readout. If you set the right eyedropper readout to RGB, (this is in the gg 2.2 document after using Apply Image), you get 12, 13, 13, 15, 16. The block that was originally 14 in the ARGB document shows up as level 13 in the gg2.2 document. But, the 32 bit readouts are identical. 16 bit has some differences between ARGB and gg2.2, even using the Apply Image (1542 > 1543, 1928 > 1927, 2056 > 2055). I suspect these have to do with the precision of the math. At least all the pixels in each block are identical.
I'll be out of pocket for the next week or two, so I may not be able to read or post much. But, it is an interesting experiment. Perhaps Apply Image is getting it right, as corroborated by the 32 bit readout, but the RGB numbers report it differently. Paste plain doesn't work right (at least for scientific purposes).
If you set the right eyedropper readout to RGB, (this is in the gg 2.2
document after using Apply Image), you get 12, 13, 13, 15, 16.
Yes, I think you are just seeing a color conversion being performed by
Photoshop from gg2.2 to aRGB for the Info Palette RGB values. We already
know that 14 goes to 13 when you color convert.
Perhaps Apply Image is getting it right, as corroborated
I've not used Apply Image before, so I don't really know anything about it.
It feels right to me though that there should be no color conversion with
this tool. I bet some wizard over at the Photoshop forum would be able to
answer this question in no time.
I use Apply Image frequently, so I am quite familiar with it. I also use calculations for some operations, and I suspect it follows suit with Apply Image as well, though I haven't tested it with this scenario.
Learning in PS is endless!
Ok, I just figured out what to do. While it DID have the same problem on my desktop computer (I was mistaken before when I said it didn'), what I found was just as important. I found how to fix it so "grayscale" mode (and individule R G and B color channels) convert PERFECTLY number-for-number into RGB mode. As I said before, on my desktop installation of PS I had more color profiles to experiment with. The perfect combination to get number-for-number exact conversion between Grayscale (or single R G and B channels via copy+paste) and RGB modes is:
Gray Profile = sGray (file = "sGray.icc")
RGB Profile = sRGB IEC61966-2.1 (file = "sRGB IEC61966-21.icc")
It is NOT effected by "use black point compensation". However it IS effected by the conversion engine you choose, and it MUST be set to "Adobe (ACE)". It won't work if you use Microsoft ICE. Also the "intent" must be set to Perseptual, Saturation, or Relative Colorimetric (which of these 3 you pick doesn't matter as it behaves the same for any of them if you use the color space combination I'm talking about). Don't pick Absolute Colorimetric, or it won't work (Gray=255 becomes R=255, G=253, B=221, instead of R=255, G=255, B=255).
I've exported the 2 icc color profile files from Photoshop and zipped them up in a file called "sRGB and sGray.zip", and uploaded to mediafire.com.
You can download it here:
I'm using Photoshop 7…
Make sure you have applied the 7.0.1 update.
Interesting. I see that sGray's kTRC is identical to all three TRCs in sRGB
which have slope limiting built into the curves...
Thanks for reporting back.