About wide vs. small gamuts, 8 vs. 16-bit &sRGB vs. Adobe'98

Started Apr 3, 2003 | Discussions
Magne Nilsen Senior Member • Posts: 1,120
About wide vs. small gamuts, 8 vs. 16-bit &sRGB vs. Adobe'98

(This is an early preview of an article that soon will appear on the http://www.etcetera.cc site)

With the many discussion appearing lately about colors, color spaces, 8-bit vs. 16-bit and sRGB vs. Adobe RGB (1998), there seem to be as many preferences as there are combinations, and most proponents of one versus the other often seems more confused than the other. I have a feeling that some perspective could be needed. It is not too hard to understand most of this, so...

This is an honest attempt to set some basic terms straight. The accompanying illustrations are made to perceptually convey what is discussed, and are not meant to be scrutinized by their byte values. They are for 'your eyes only'.

Let's simplify a little. The example above shows two gamuts. A gamut is all the colors that can be reproduced or captured by a device. The number of bits used to hold the gamut dictates the number of steps that each color can contain. Many bits will give room for many subtle variations or tones per color. The width of the gamut puts constraints upon, or limits the saturation level of colors inside the gamut. A wide gamut can hold very saturated colors, while a small gamut will sacrifice those most vivid colors. As long as the gamut is, and stays, both wide and has many bits to represent the color values, few problems should occur in handling images. As you see from the illustration above, the "large gamut" version has more saturated blues, greens, reds, yellows etc. Why would anyone consider the "small gamut" version at all?

These are the same gamuts as in the first image, but this time represented with fewer bits. The large gamut version still has more saturated colors, but now there is a price to pay! The "few-bits large gamut" has sacrificed many of the more subtle tonalities to give room for the more saturated and neon-like colors. Which finite set of the colors above do you think would be the best to accurately describe human skin, like in a portrait? Or - a sunset? Or - a small lake with a green forest around?

These illustrations are much worse than the difference between 8-bit vs. 16-bit, and sRGB vs. Adobe RGB (1998), but this is still the dilemma you are going to face in the real world of printers and output devices. Sooner or later.

At some stage the "many-bits large gamut" image must be color-wise transformed into a "not-so-wide" and "not-so-many-bits" gamut, representing a display, a web page or a printer. Printer gamuts are actually mostly smaller than sRGB.

At some stage you will be doing things in 8-bit software as well. Photoshop has some limitations on the available functionality for 16-bit images, but I would still advice all to switch to 16-bit as soon as possible, and stay there as long as possible when editing images. When you go from “wide gamut 16-bit” to “small gamut 8-bit” you should know what will happen, or you could be in for a surprise, or even worse, your customer/client/uncle could be in for a color-wise surprise.

As an example on how bad things can get, look at the last illustration.

Applying an S-curve as strong as the one above onto the prior shown “few-bits large gamut” image will result in a reduction from 400 finite colors down to 300. 100 colors (25% of the gamut!) will be gone, and many colors will show up as identical in the image. The same thing would happen if we were to show a “few-bits huge gamut” image on your “not so huge gamut” display. You would not be able to discern many of the colors from each other at all.

My intention with this is not to scare anyone away from using a wide gamut working space to hold or edit their images. My advice would be to check up what exact colors you are getting in Adobe RGB 1998 that are not present in sRGB and vice versa. I think you could be surprised.

16-bit editing is almost without an exception an advantage. Especially if you are going to use a wider gamut like Adobe 1998 RGB (or maybe even wider?). You will have a much bigger latitude in you editing operations before you start to experience ugly 8-bit problems like banding of neighboring colors, or posterization problems in the shadows or in skies. Switching to, and staying in 16-bit as long as possible, is almost always an advantage for your images. I think 16-bit is much more important than using Adobe RGB 1998 as a working space, but I constantly find professionals that have heard that Adobe RGB 1998 is so good, but they still do all their editing in 8-bit, and wonder why they have banding and posterization problems. Most of them believe that this would be even worse if they had sRGB as their working space, while the opposite is true. I hope that became clear with the prior illustrations.

Ah well, 16-bit is “better” than 8-bit then, but wide gamut is not “better” than small gamut, although the words wide and small for many would indicate so. It really depends. On the destination target, on the software available, on your skills and knowledge, on your input device/monitor/output devices level of quality, calibration and exactness, and on a few physical truths like bandwidth, memory, processor and hard disk size and quality. In many cases it is actually an advantage to be working in a “not-so-wide” gamut like sRGB. You get more subtle transitions at the cost of some “neons”, and you will be much less surprised or disappointed when transferring your images to the web, the printer, the lab, the clients or your friends.

~~~~
Magne

Mishkin Veteran Member • Posts: 3,654
Thanks for the primer, Magne, and one gripe:

Photoshop's 16-bit SUCKS!!! You can't have layers, you can do almost nothing but some color adjustments.... Ugh!!!

Magne Nilsen wrote:

(This is an early preview of an article that soon will appear on
the http://www.etcetera.cc site)

DanSan Senior Member • Posts: 1,145
Thanks. very helpful (nt)
-- hide signature --

'We demand rigidly defined areas of doubt and uncertainty' HHGTTG
http://www.pbase.com/dasantillo

Jason Hutchinson Forum Pro • Posts: 12,177
I've tried to tell people this from time to time

Magne Nilsen wrote:

Ah well, 16-bit is “better” than 8-bit then, but wide gamut is not
“better” than small gamut, although the words wide and small for
many would indicate so. It really depends. On the destination
target, on the software available, on your skills and knowledge, on
your input device/monitor/output devices level of quality,
calibration and exactness, and on a few physical truths like
bandwidth, memory, processor and hard disk size and quality. In
many cases it is actually an advantage to be working in a
“not-so-wide” gamut like sRGB. You get more subtle transitions at
the cost of some “neons”, and you will be much less surprised or
disappointed when transferring your images to the web, the printer,
the lab, the clients or your friends.

...especially those that say Adobe RGB will automatically get them "smoother colors." Good to hear some info from someone who is up on this stuff.

Jason

OP Magne Nilsen Senior Member • Posts: 1,120
Photoshop 16-bit...

Mishkin wrote:

Photoshop's 16-bit SUCKS!!! You can't have layers, you can do
almost nothing but some color adjustments.... Ugh!!!

It obviously takes Adobe a lot of time to enable 16-bit for a wider set of the functions, but at least some of my favorites are in place. For full image editing Curves and Levels are fully there, and I tend to do > 2/3 of my work in Curves. Segmented editing seems worse unless you can live with Marquees and Lassos to make selections. I tend to make a 8-bit copy of my 16-bit image, create and then Save my selections there, and then from the 16-bit image use "Load Selection" from the 8-bit image. Works fine for me. As for the filters, well there is Gaussian Blur, Median and Unsharp Mask. Hmmm, that is about the only filters I use for image correction and editing, so I don't really care. If you must have the rest of the filters, you have to switch into 8-bit - yes, but I find that most of them are more on the artistic side, and seldom use them. What I really miss is cut & paste, layers and transform...

~~~~
Magne

Steve Deutsch Contributing Member • Posts: 948
Very nice. But what does...

The last piece of the puzzle for me is this:

What does it mean to use Photoshop (example) in a given workspace?

Seems to me that math is math. 16 is better math than 8. Agreed.

Why should I care what colorspace I work in? Yes I want a good mapping from my 'internal numbers' when I generate the displayable output to an sRGB or other colorspace device...but I don't see how the colorspace of my output devices means anything to Photoshop until I generate displayable results.

What's up with this?

Khanh M Senior Member • Posts: 1,880
Re: About wide vs. small gamuts, 8 vs. 16-bit &sRGB vs. Adobe'98

Magne Nilsen wrote:

(This is an early preview of an article that soon will appear on
the http://www.etcetera.cc site)

OP Magne Nilsen Senior Member • Posts: 1,120
Re: Very nice. But what does...

Steve Deutsch wrote:

The last piece of the puzzle for me is this:
What does it mean to use Photoshop (example) in a given workspace?

Not sure I know what your question is here, but you select your working space from the "> Edit/Color Settings" dialog. If you select a working space different from most of your images gamut, you must be prepared to understand and use the functions for Assigning and Converting images between different colorspaces. Ian Lyons has many great tutorials i.e. here: http://www.computer-darkroom.com/ps7-colour/ps7_1.htm

Why should I care what colorspace I work in? Yes I want a good
mapping from my 'internal numbers' when I generate the displayable
output to an sRGB or other colorspace device...but I don't see how
the colorspace of my output devices means anything to Photoshop
until I generate displayable results.
What's up with this?

If your working colorspace is very different from your output devices some conversion will occur. Many of those conversion will mean that colors in your images will perceptually change. Sometimes a lot. Unless you have a VERY good profile for that output device, and know exactely how to do and understand softproofing, what you see on your screen can be very far from what you will see coming out of your output device. Generally not a good thing. And - if your working space is far from sRGB you must always convert images to sRGB when sending them out to the mostly non-color-managed world outside of Photoshop...
I hope this was relevant to your question...

~~~~
Magne

Steve Deutsch Contributing Member • Posts: 948
Re: Very nice. But what does...

Thanks, but let me ask it this way.

If I set the Photoshop 'Working Spaces' RGB setting to AdobeRGB, and my monitor and printer are sRGB...what have I gained? Wouldn't a working space of sRGB do just as good, maybe better because the mapping to the monitor or printer ICC profile isn't much of a change?

Magne Nilsen wrote:

Steve Deutsch wrote:

The last piece of the puzzle for me is this:
What does it mean to use Photoshop (example) in a given workspace?

Not sure I know what your question is here, but you select your
working space from the "> Edit/Color Settings" dialog. If you select
a working space different from most of your images gamut, you must
be prepared to understand and use the functions for Assigning and
Converting images between different colorspaces. Ian Lyons has
many great tutorials i.e. here:
http://www.computer-darkroom.com/ps7-colour/ps7_1.htm

Why should I care what colorspace I work in? Yes I want a good
mapping from my 'internal numbers' when I generate the displayable
output to an sRGB or other colorspace device...but I don't see how
the colorspace of my output devices means anything to Photoshop
until I generate displayable results.
What's up with this?

If your working colorspace is very different from your output
devices some conversion will occur. Many of those conversion will
mean that colors in your images will perceptually change. Sometimes
a lot. Unless you have a VERY good profile for that output device,
and know exactely how to do and understand softproofing, what you
see on your screen can be very far from what you will see coming
out of your output device. Generally not a good thing. And - if
your working space is far from sRGB you must always convert images
to sRGB when sending them out to the mostly non-color-managed world
outside of Photoshop...
I hope this was relevant to your question...

~~~~
Magne

OP Magne Nilsen Senior Member • Posts: 1,120
Re: Very nice. But what does...

Steve Deutsch wrote:

Thanks, but let me ask it this way.
If I set the Photoshop 'Working Spaces' RGB setting to AdobeRGB,
and my monitor and printer are sRGB...what have I gained? Wouldn't
a working space of sRGB do just as good, maybe better because the
mapping to the monitor or printer ICC profile isn't much of a change?

Any advandages of a wider gamut like AdobeRGB would be for editing latitude of the most vivid colors. Less color clipping could occur in long sequences or during heavy editing. The posibillities of those quantization and rounding errors can be less with a wider space if you solely think about the more saturated than sRGB colors. But, it can get worse for your "inside sRGB gamut" colors since there are fewer "buckets" to hold the sRGB part of the gamut. If everything happened in 16-bit you would be safe, but even with the smallest editing in 8-bit you would be worse off with AdobeRGB than with sRGB, IMHO. For images, I think that you - as most others - are best adviced to use sRGB only. There are situations that demands a larger gamut, but most of them are outside the photgraphic world, like in creating logos, art, etc. that have a fair chance of beeing reproduced on relevant output equipment.

~~~~
Magne

alFR Regular Member • Posts: 286
Re: Very nice. But what does...

I understood that from version 7, if you opened a document with a colour profile embedded Photoshop used that colourspace as the working space e.g. open a sRGB-tagged image, you edit in sRGB. I thought that the working space set within PS only determined the colourspace for new documents or those without an assigned profile. Is that correct?

As an aside, there was a big debate over at Robgalbraith.com about whether to convert documents to your working space or not, i.e. if you open a D60 image that has the sRGB (or a custom) profile assigned, should you convert it to AdobeRGB for editing or just leave it. Ian Lyons was of the opinion that you should just leave it, as the conversion to AdobeRGB wouldn't yield any real-world benefits and might degrade the image slightly. His argument seemed logical and now I've stopped converting to AdobeRGB as part of my workflow. Comments anyone?

Magne Nilsen wrote:

Any advandages of a wider gamut like AdobeRGB would be for editing
latitude of the most vivid colors. Less color clipping could occur
in long sequences or during heavy editing. The posibillities of
those quantization and rounding errors can be less with a wider
space if you solely think about the more saturated than sRGB
colors. But, it can get worse for your "inside sRGB gamut" colors
since there are fewer "buckets" to hold the sRGB part of the gamut.
If everything happened in 16-bit you would be safe, but even with
the smallest editing in 8-bit you would be worse off with AdobeRGB
than with sRGB, IMHO. For images, I think that you - as most others

  • are best adviced to use sRGB only. There are situations that

demands a larger gamut, but most of them are outside the
photgraphic world, like in creating logos, art, etc. that have a
fair chance of beeing reproduced on relevant output equipment.

~~~~
Magne

OP Magne Nilsen Senior Member • Posts: 1,120
Re: Very nice. But what does...

alFR wrote:

I understood that from version 7, if you opened a document with a
colour profile embedded Photoshop used that colourspace as the
working space e.g. open a sRGB-tagged image, you edit in sRGB. I
thought that the working space set within PS only determined the
colourspace for new documents or those without an assigned profile.
Is that correct?

Generally it will "Use the embedded profile" if you say so, but this all depends on, and is controlled by your settings inside PhotoShops Edit-Color Settings:

and your selected Working Space in the same dialog.

As an aside, there was a big debate over at Robgalbraith.com about
whether to convert documents to your working space or not, i.e. if
you open a D60 image that has the sRGB (or a custom) profile
assigned, should you convert it to AdobeRGB for editing or just
leave it. Ian Lyons was of the opinion that you should just leave
it, as the conversion to AdobeRGB wouldn't yield any real-world
benefits and might degrade the image slightly. His argument seemed
logical and now I've stopped converting to AdobeRGB as part of my
workflow. Comments anyone?

I am very much on Ian's side in this. Any color space conversions has an element of rounding errors and quantization problems in them, and as long the original color space is a real working space with the same gamma values (2.2) - like sRGB or AdobeRGB - and not a device profile space (like a camera/scanner gamut) I would advice to stay in the original color space.

A typical "convert to profile" conversion means going "down" to a gamma 1.0 wide Lab space - and then back to a gammacorrected less-wide space again. Adobe has stated that those conversion are always done in 16-bit no matter if the image is 8-bit, but a typical down to gamma 1.0 and back to gamma 2.2 is still a big conversion...

The critical aspect is often to know when "real-world benefits" occurs and when "real-world problems" occurs. Many of the AdobeRGB (and even wider gamuts) proponents often wants a simple answer for a complex question. Unfortunately, real life (or politics) often doesn't behave like that.

~~~~
Magne

alFR Regular Member • Posts: 286
Thanks. One more question...

Magne Nilsen wrote:

I am very much on Ian's side in this. Any color space conversions
has an element of rounding errors and quantization problems in
them, and as long the original color space is a real working space
with the same gamma values (2.2) - like sRGB or AdobeRGB - and not
a device profile space (like a camera/scanner gamut) I would advice
to stay in the original color space.

Thanks for that. In view of what you say about quantization errors in colourspace conversions, what do you think is the best workflow out of these?

1. Open D60 image in PS using default sRGB space (based on the EXIF colourspace tag), edit, print.

2. Open D60 image in PS, assign custom camera profile, convert to working space, edit, print.

i.e. do the errors introduced by the conversion to the working space negate the benefits of assigning a custom profile?

Al

OP Magne Nilsen Senior Member • Posts: 1,120
Re: Thanks. One more question...

alFR wrote:

In view of what you say about quantization errors
in colourspace conversions, what do you think is the best workflow
out of these?
1. Open D60 image in PS using default sRGB space (based on the EXIF
colourspace tag), edit, print.
2. Open D60 image in PS, assign custom camera profile, convert to
working space, edit, print.

For the D60 clearly #2, since the results of Canons conversion IMO is relatively too far away from sRGB - colorwise. For other cameras it depends on how good job the manufacturer have done in delivering an image conversion close enough to sRGB or Adobe'98 - i.e. the 10D looks much closer to accomplishing just that.

i.e. do the errors introduced by the conversion to the working
space negate the benefits of assigning a custom profile?

No - if the color values are clearly off you are best served with an input device profile. Preferably from a 16-bit linear image since that would give a much smoother conversion. With a already gammacorrected image the conversion must be done in the form of a huge LUT (Look-Up-Table) that essentially undos the manufacturers errors Some of those undos can be a bit on the harsh side, but should not normally incur any visible problems.

The rounding and quantization errors that occurs in these colourspace conversions would not normally accumulate to more than 1 or maximum 2/255 of a R,G or B value change. I have not met any humans that can perceptually catch that small changes yet, but if this occurs in the shadows and you later bring those shadows up a lot with curves, you could probably see some of this as posterization or banding, but I think you had to work hard to make this a problem

~~~~
Magne

Steve Deutsch Contributing Member • Posts: 948
What's 'the best' D60 ICC

Aside from the fact that 'the best' is subjective....what is the best D60 ICC?

Magne Nilsen wrote:

alFR wrote:

In view of what you say about quantization errors
in colourspace conversions, what do you think is the best workflow
out of these?
1. Open D60 image in PS using default sRGB space (based on the EXIF
colourspace tag), edit, print.
2. Open D60 image in PS, assign custom camera profile, convert to
working space, edit, print.

For the D60 clearly #2, since the results of Canons conversion IMO
is relatively too far away from sRGB - colorwise. For other cameras
it depends on how good job the manufacturer have done in delivering
an image conversion close enough to sRGB or Adobe'98 - i.e. the 10D
looks much closer to accomplishing just that.

i.e. do the errors introduced by the conversion to the working
space negate the benefits of assigning a custom profile?

No - if the color values are clearly off you are best served with
an input device profile. Preferably from a 16-bit linear image
since that would give a much smoother conversion. With a already
gammacorrected image the conversion must be done in the form of a
huge LUT (Look-Up-Table) that essentially undos the manufacturers
errors Some of those undos can be a bit on the harsh side, but
should not normally incur any visible problems.

The rounding and quantization errors that occurs in these
colourspace conversions would not normally accumulate to more than
1 or maximum 2/255 of a R,G or B value change. I have not met any
humans that can perceptually catch that small changes yet, but if
this occurs in the shadows and you later bring those shadows up a
lot with curves, you could probably see some of this as
posterization or banding, but I think you had to work hard to make
this a problem

~~~~
Magne

alFR Regular Member • Posts: 286
Re: Thanks. One more question...

Thanks for the advice.

Any sign of your D60 profiles appearing in the near future?

OP Magne Nilsen Senior Member • Posts: 1,120
About D60 ICC profiles and subjectivity...

Steve Deutsch wrote:

Aside from the fact that 'the best' is subjective....what is the
best D60 ICC?

Aside from the fact that some are a tad more subjective than others:

  • for D60 linear images: IMHO, the "D60 ETC-0" profile

  • for D60 non-linear images; direct in-camera JPGs and nonlinear JPG or TIFF images converted with the Canon SDK based software, including Canons RIC and FVU, Breezebrowser and YarcPlus: IMHO, the "D60 ETC-3" profile

These two profiles have been in beta testing for many months and are now finalized. The release of both these profile was delayed some weeks, while awaiting a strategically confirmation that would have meant an other form for distribution than originally planned. These discussions have now come to an end, and the original plan will be fullfilled ASAP. Hopefully should the http://www.etcetera.cc site be opening later this week. The profiles will not be free, I prefer to call it "a handlling fee", of $25 each, or $40 if both are bought together. There are already > 2000 names on the information list at info@etcetera.cc and by sending a mail to that list stating that you want information, you will also be automagically be reminded when the site opens. The etcetera site will hopefully become an "icc profile foundry" and I have already started the profiling of 4 other cameras...

Since I personally made both these profiles my degree of "subjectivity" should not be underestimated, and hopefully others will be able to fill in about the profiles quality.

For samples and images where these and other profiles have been applied, please look at this thread:
http://forums.dpreview.com/forums/read.asp?forum=1019&message=4608863

~~~~
Magne

Leon Wittwer Forum Pro • Posts: 13,204
Re: Very nice. But what does...

Magne Nilsen wrote:

I am very much on Ian's side in this. Any color space conversions
has an element of rounding errors and quantization problems in
them, and as long the original color space is a real working space
with the same gamma values (2.2) - like sRGB or AdobeRGB - and not
a device profile space (like a camera/scanner gamut) I would advice
to stay in the original color space.

The one exception is where you start with an sRGB image and want to extend the saturation. Then it is usually better to convert to Adobe RGB (1998).

~~~~
Magne

Leon Wittwer Forum Pro • Posts: 13,204
Re: About wide vs. small gamuts, 8 vs. 16-bit &sRGB vs. Adobe'98

Magne, After reading your writing and all of the messages I would add another comment. People sometimes should rely less on general rules than on looking at the scene and figuring out what is best. Depending on the image content, I might go either with sRGB or Adobe. From most of my landscape work in the southwest US, sRGB is best as there don't tend to be colors that push the gamut. As you note, I then get the advantage of better tonal resolution. In other cases, like doing flowers with green leaves, then Adobe RGB (1998) is usually best. When I am not sure, I tend towards Adobe RGB (1998) to cover my bets and then use Photoshop adjustments and rendering intent to go to the eventual smaller print space. Of course, this all is a great argument for doing raw as you can defer the decision until you can see the image on a monitor. I did enjoy this and your other contributions. Leon

Magne Nilsen wrote:

(This is an early preview of an article that soon will appear on
the http://www.etcetera.cc site)

~~~~
Magne

Leon Wittwer Forum Pro • Posts: 13,204
Re: About wide vs. small gamuts, 8 vs. 16-bit &sRGB vs. Adobe'98

I would add to my earlier comments that it is good to minimize the number of operations to minimize, in turn, quantization effects like banding etc. Repeatedly reworking an image will likely cause problems. Using adjustment layers is a good way to store sets of changes without permanently messing up an image. If you need to rework something, you just go back as far as you can and do the fix. Leon

Keyboard shortcuts:
FForum MMy threads