Bit Depth and Color Spaces

[tthaley]

The "Influence on Bit Depth Distribution" section of the "SRGB vs. Adobe RGB 1998" tutorial seems to contradict other information found on this site.

Since humans can only perceive about 10 million colors, isn't 24 bpp (16777216 colors) a good fit for a color space like ProPhoto RGB, which approaches the range of all colors visible to humans? Given this, why would it be necessary to restrict Adobe RGB to 16 bpc images instead of 8 bpc (24 bpp) images? Isn't 8 bpc good enough for all perceivable color spaces?

[george013]

I think the value is in editing. Editing with a higher bitdepth gives a smoother change of luminance, colour.
George

[tthaley]

I agree that higher bit depth is helpful for preserving detail in post-processing. But can a final image be saved in a large color space (like ProPhoto) at 8 bpc without any loss of detail vs. using, say, 16 bpc with that same large color space?

[Manfred M]

Tom - this is a common mistake (and one I made too). You are making the incorrect assumption that the colour spaces in question cover what human eyes can see. This is not correct. Humans see more colours, especially in the blues and greens thatn sRGB can reproduce. This also goes the other way, certain colour spaces can represent colours outside of the human visual range; ProPhoto is a common example of this. The two most common colour spaces sRGB 1996 and AdobeRGB were developed around two very specific technical issues. The sRGB was written around the colour range the computer screen was capable of displaying and AdobeRGB was an attempt to unify the colour mapping of screen to printer; i.e. a way of describing a colour that is going to give you the same result, regardless of whether you are using a printer or computer screen.

A common analogy of colour spaces is a ladder, where the height of the ladder is a specific colour reprentation, and the bit-depth represents the height of each run on the ladder. An 8-bit ladder would have very large spacing between the rungs, whereas a 16-bit ladder would have very little distance between rungs. So long as the ladder is as long as or longer than the height you are trying to get to, it's fine, but if it is too short, you can't get there.

When we edit an image, we are effectively remapping colour values. If there are too few of them and we make signfiicant enough changes, we will start noticing this. Using a broad colour space like ProPhoto and a 16-bit representation of the colours reduces the risk of introducing these noticable issues (i.e. "artifacts") in our edits.

If we downgrade out bit-depth and colour space as our very last step in the editing process. This is in fact what we are doing when we output a 16-bit AdobeRGB or ProPhoto colour space image to an 8-bit sRGB jpeg for posting on the internet. Minimal risk of artifacts in that workflow.

[ajohnw]

The 16 bit Adobe RGB seems to be cropping up again. It's basically 3 8 bit colour channels just like sRGB. aRGB just spans a greater colour space with bigger gaps and was designed to maximise the capabilities of CMY printing. Both aRGB and sRGB can display colours that the other can't. This is rather important when it comes to displaying on a PC screen. Some people like to force aRGB onto sRGB screens but the result is often a bit Constable painting like - the way I put it. Flatter and more subdued might be another way of putting it. It's just a case of adjusting an aRGB image on a sRGB monitor. Some people post aRGB shots on the web even though there are more sRGB screens about. I believe it's also possible to adjust so that there will be little difference.

Keeping a deeper colour space about while editing images is more important but as many GIMP users would point out when it just retained 8 bit not as dramatic as might be expected. The main problem can be banding when tones are stretched too much - often only needed on a badly exposed or poor problematic choice of shooting conditions. The GIMP doesn't process raw images anyway. That is done with full bit depth software. Do that badly and banding is very likely following subsequent PP. And actually the GIMP still discards information from some higher bit depth formats when they are loaded.

The number of colours the human eye can see is very debatable especially when it comes to shades and brightness levels. It's an analogue system without any steps. Gamuts are nearly always shown in 2D. Really they need to be shown in 3D to see differences. One of the problems with achieving the same gamut is that it can't be spanned by 3 colour mixing without introducing imaginary colours. Imaginary colours are ones that we can't see - fine in software but not of much use when it comes to viewing on a PC screen. Actually despite claims I don't think ProColour printing really covers the entire human eye gamut. There is also the problem that cameras at some single exposure setting just don't have as much dynamic ranges as the colour vision of out eyes has. PC screens and printers are much much worse which is the real problem.

John
-

[LocalHero1953]

The 16 bit Adobe RGB seems to be cropping up again. It's basically 3 8 bit colour channels just like sRGB.
-

Are you sure? What I thought I thought was the case, and Manfred seems to say the same, is that the colour space is a separate issue from the bit depth. So one could have 16bit sRGB and 8bit Adobe RGB if one wanted. Typically, though, I post process in 16bit in Adobe RGB (starting from raw in Lightroom), and then export in sRGB for the web as JPGs, which are by definition only 8bit.

Without ever having done any tests, I have always assumed the benefit of 16bit to be smoother tonal and colour gradations.

(Edited for clarity)

[Manfred M]

The 16 bit Adobe RGB seems to be cropping up again. It's basically 3 8 bit colour channels just like sRGB.

Strangely enough, I would have to disagree. I'm with Paul (and Adobe) on this one. Colour space and bit depth are independent, When I use Adobe Camera RAW I can chose to work in a multitude of different colour settings and all appear to be able to work in 8-bit and 16-bit. Both sRGB and AdobeRGB are available in 8-bit and 16-bit configuration.

Yes, I did check things just to make sure.

[ajohnw]

Afraid in practical terms - viewing - I'm going to stick with it. Some time ago I did assume that aRGB had more colours where as it has exactly the same number of colours as sRGB. Part of my confusion was caused by the fact that aRGB monitors use 10 bit panels ideally but usually achieving the 10 by 8 + 2 bit dithering. The reason for the need for 10 is the colour span of the gamut.

When it comes to works spaces things are entirely different. I can for instance choose prophoto while producing sRGB images. actually that packages states that I should. Also a 16 bit one or even jpg / 8 bit.. 2 of the packages I use a lot use 32 bit floating point. Another standard format with way more dynamic range than any of the others just mentioned. That has nothing to do with viewing only how the data is retained.

If some one wants to say that there are other for instance jpg standards in term of how the actual image is stored but pass on how they can be viewed - yes there are. I saw mention of a 12 bit one recently. Yes jpg's. There are lots of standards.

https://www.cambridgeincolour.com/tu...obeRGB1998.htm

http://www.kenrockwell.com/tech/adobe-rgb.htm

http://petapixel.com/2009/09/17/why-...ably-use-srgb/

I could go on.

John
-

[Manfred M]

Afraid in practical terms - viewing - I'm going to stick with it.

Afraid I am going to have to disagree with you again. You are being rather misleading here. I will have to totally agree with you that given the same bit depth, both sRGB and Adobe RGB are capable of reproducing exactly the same number of colours; 2^8 and 2^16 respectively. What you have carefully chosen to disregard is that sRGB covers roughly 35% of the visible colours, while Adobe RGB covers just over 50% of the visible colours. This is of both real and practical importance; the steps between discrete shades in AdobeRGB are larger than the ones in sRGB.

I agree that computer screens use a 8+2 model, where you have a native 8 bits per channel and the other two bits are generated by dithering; these displays use a variant of IPS technology. While it is quite possible that there are true 24-bit sRGB displays out there the bulk of the lower end displays use TN technology which has a native 6-bit resolution and the other two bits are dithered (6+2). I'm staring at one example each computer screen type of each as I type this; my main screen is a high end one and is rated at around 96% Adobe RGB compliant and 100% sRGB compliant and the second one (where I park my menus and secondary operations) is 100% sRGB compliant. Can I see the difference in colours between the two colour spaces on the two screens; ABSOLUTELY! The wide gamut screen is more accurate when I set myself to use sRGB (and yes both have been calibrated and profiled).

As for using a wide gamut colour space for editing; yes, there I do agree with you. However, regardless of the different jpeg standards; the 8-bit one is the one that is used on the internet, not the exotic variants that virtually nobody uses.

[ajohnw]

No Manfred I am not ignoring that - I pointed out that the aRGB has a wider gamut but in terms of I just like others feel that the gaps are important when it comes down to viewing on a PC screen. Printing, which is what aRGB is aimed at is an entirely different matter and also has a lower final dynamic range. If people look for long enough there are web pages about that show the differences these gaps produce.

As far as screens go the panels in sRGB screens are largely dependent on price paid. There have been and still may be 6+2 screens about. TN not so sure. The last TN screen I had was bought an awful long time ago. It wouldn't surprise me if no one makes them any more. I started using them shortly after they came out and have also use true LCD screens on laptops - awful things. TFT was a huge improvement.

There is a good description of gamut here for any that are interested. It also mentions another point I have mentioned before about the differences between an aRGB and an sRGB screen -

(LCD) screens filter the light emitted by a backlight. The gamut of an LCD screen is therefore limited to the emitted spectrum of the backlight. Typical LCD screens use cold-cathode fluorescent bulbs (CCFLs) for backlights. LCD Screens with certain LED or wide-gamut CCFL backlights yield a more comprehensive gamut than CRTs. However, some LCD technologies vary the color presented by viewing angle. In Plane Switching screens have a wider span of colors than Twisted Nematic.

http://en.wikipedia.org/wiki/Gamut

My avoidance and intense dislike of the use of the term LCD comes from having to use one but I have to put up with others using it.

There are other factors in this area and I wish I had kept a link. One a test on a camera. Shot taken of a standard test card. Shown with smaller squares in each one that were the actual colours the camera produced. Very obvious visual errors. There is an Australian test site that always tests cameras this way but presents them as gamut point plots which doesn't really show how bad the error is. Basically cameras are not that accurate anyway. Another pointed out that a recent high fashion colour couldn't be reproduced accurately. I assume they mean with more normal camera reproduction methods which I would assume includes aRGB. Maybe it's like that teal colour that causes sRGB problems. In practice probably the only one that causes real problems.

My screen

The Dell U2713HM utilises an LG.Display LM270WQ1-SLB2 AH-IPS panel which is capable of producing 16.7 million colours with a true 8-bit colour depth. Dell refer to the panel as being "AH-IPS" (Advanced High Performance IPS) in some of their marketing material, including the original Japanese press release, and it is a name which is starting to become more common in today's IPS market. We have started to see other screens emerge with these so-called AH-IPS panels in their specs, and indeed LG.Display themselves made the

There are only a very limited number of actual panel makers.

John
-

[tthaley]

From the discussion here it seems that gamut describes the range of colors, whereas bit depth describes the number of colors available. Furthermore, it seems that a gamut that includes 100% of the colors visible to humans would be best if viewed at 8 bits (no dithering), i.e. any increase in bit depth does not help with detail or color accuracy. If this is true, then why do displays with 12-bit color exist?

So 16-bit and higher bit depth is only useful for preserving details during post-processing.

One more thing about color spaces... I recently watched a tutorial about color spaces (as part of a printing series on lynda.com) and the instructor said that Adobe RGB is more true-to-life than ProPhoto:

Why then don't I stay in ProPhoto all the time? In general, taking your colors and mapping them across that big a color space can sometimes produce problems. It can cause a color shift, your yellows might go more in one direction or another.

It can also cause banding and skies or gradients places where it is having to stretch color across a broader space.

I would tend to disagree, since ProPhoto encompasses Adobe RGB and (maybe?) the colors that can be captured by the camera. It is more than likely that Adobe RGB is the less accurate of the two, since colors that exist in ProPhoto cannot exist in Adobe RGB. Of course, no monitor can display ProPhoto, but I'm not sure what role this plays. I also don't understand why colors would be "stretched" into a color space if the source space is within the destination space.

[Polar01]

Tom this colour space things goes round and round. Lets say all you want to do is post images to the web or put them on a stick or card and show them on a TV or someone's computer or maybe print on a bookshelf CMYB printer, than all you ever need is a jpeg which by way is sRGB and 8-bit. If you sent your image out to most photo labs all they want is sRGB any baybe 240 PPI instead of the 72 that is used for web. If you plan on some custom inkjet printing (printers can have up to 11 or more inks) then it is best to work in the largest colour space ProPhoto save in that and let the printer take care of it.
Life for some of us is to short to read spec after spec after spec, instead just take the shot and make the best image possible.

Cheers: Allan

[ajohnw]

Tom - Show me a 12bit colour display. I assume you mean 10 bit.

Your best bet for getting an understanding is to read the CinC tutorial I posted a link to earlier very carefully.

What causes a lot of confusion in most cases in this area is that when we are using a PC monitor we are working in 3 x 8 bit colour spaces visually. The extra bit resolution available underneath is essentially fractions of these 8 bits and can not be seen visually. PP can be used to make these fractions visible. This is why some say working in 8bit throws information away because these fractions are being truncated in some way. Taking a simple example 12bit has 16 times the resolution of 8 bit so each of the bits in 8bit has 1/16 bits "underneath". This is why working spaces generally have higher bit depths than output formats / colour spaces. There is another aspect as well. Say some one is working in 8bits 0 to 255 and and pushes something up to 256 - it's gone and can't be restored because it's outside the scope of the number range. If the number range is larger it can be bought back again.

People also work in ProPhoto for printing. In this case they are working on colours that they can not visually see so have to take care of that some other way. Your instructors comment doesn't make any sense at all This describes prophoto and shows that in practice speaking literally people can not generate all of the output colours that it contains as some are imaginary.

http://en.wikipedia.org/wiki/ProPhoto_RGB_color_space

John
-

[tthaley]

Is this not a 12 bit display? I noticed the phrase "68.7 Billion Colors (99% of Adobe RGB)" on this page is misleading, as it seems to indicate that more colors directly equates to a larger color space.

[ajohnw]

Is this not a 12 bit display? I noticed the phrase "68.7 Billion Colors (99% of Adobe RGB)" on this page is misleading, as it seems to indicate that more colors directly equates to a larger color space.

Things are expressed more clearly hear. The 14bit LUT is is used to correct the colour response of the monitor. Remember that the 10 bit aspect is needed to span the aRGB gamut and doesn't signify numerically more colours just a wider range tha sRGB.

http://www.tftcentral.co.uk/reviews/...nic_vp2772.htm

Later on in the write up it mentions higher bit depth processing. This relates to processing the colours not the colour depth.

John
-

[george013]

Tom - Show me a 12bit colour display. I assume you mean 10 bit.

Your best bet for getting an understanding is to read the CinC tutorial I posted a link to earlier very carefully.

What causes a lot of confusion in most cases in this area is that when we are using a PC monitor we are working in 3 x 8 bit colour spaces visually. The extra bit resolution available underneath is essentially fractions of these 8 bits and can not be seen visually. PP can be used to make these fractions visible. This is why some say working in 8bit throws information away because these fractions are being truncated in some way. Taking a simple example 12bit has 16 times the resolution of 8 bit so each of the bits in 8bit has 1/16 bits "underneath". This is why working spaces generally have higher bit depths than output formats / colour spaces. There is another aspect as well. Say some one is working in 8bits 0 to 255 and and pushes something up to 256 - it's gone and can't be restored because it's outside the scope of the number range. If the number range is larger it can be bought back again.

People also work in ProPhoto for printing. In this case they are working on colours that they can not visually see so have to take care of that some other way. Your instructors comment doesn't make any sense at all This describes prophoto and shows that in practice speaking literally people can not generate all of the output colours that it contains as some are imaginary.

http://en.wikipedia.org/wiki/ProPhoto_RGB_color_space

John
-

John, Can you explain me what you mean with that? The maximum value in a 8-bit division is 255, in a 12-bit division it is 4096. If I go over the maximum of 255 in a 8-bit division it's the same as going over the maximum of 4096 in a 12-bit division. Both are clipping.

And I think there is a difference between the software and the hardware.

George

[Manfred M]

From the discussion here it seems that gamut describes the range of colors, whereas bit depth describes the number of colors available.

Correct.

Furthermore, it seems that a gamut that includes 100% of the colors visible to humans would be best if viewed at 8 bits (no dithering), i.e. any increase in bit depth does not help with detail or color accuracy. If this is true, then why do displays with 12-bit color exist?

Because this is NOT correct. Just as an aside, sRGB contains around 35% of the colours visible to humans, AdobeRGB contains slightly more than 50% of the colours visible to a human and ProPhoto includes some “synthetic” colours, i.e. ones that are not visible to humans. ProPhoto can render 100% of real world colours (as per the Pointer Gamut), but includes 13% imaginary colours as well.

As someone who regularly works with an 8-bit and 10-bit display (side by side), I would not agree with this. There is a difference. As to how well dithering works, I’ll leave that to the academics, but I don’t think anyone will argue that it will display more discrete colours than an 8-bit, non-dithered unit (actually, I’m sure that someone will).

So 16-bit and higher bit depth is only useful for preserving details during post-processing.

NO, as stated above.

One more thing about color spaces... I recently watched a tutorial about color spaces (as part of a printing series on lynda.com) and the instructor said that Adobe RGB is more true-to-life than ProPhoto:

Yes, but only because ProPhoto has some imaginary colours, as noted above. Technically, I would agree with the instructor just on that basis, but he also discounts the fact that AdobeRGB cannot handle almost half of the colours we humans can see. From a practical standpoint, my default editing colour space is ProPhoto.

I would tend to disagree, since ProPhoto encompasses Adobe RGB and (maybe?) the colors that can be captured by the camera. It is more than likely that Adobe RGB is the less accurate of the two, since colors that exist in ProPhoto cannot exist in Adobe RGB. Of course, no monitor can display ProPhoto, but I'm not sure what role this plays. I also don't understand why colors would be "stretched" into a color space if the source space is within the destination space.

The “stretching” into the colour space is a result of the interpolation algorithms required to map one colour space into another. Colour spaces are really 3-D models and the relevant data points will not be identical across the colour spaces, hence the need to mapping.

[LocalHero1953]

when we are using a PC monitor we are working in 3 x 8 bit colour spaces visually.

I do think we need to be rigorous in how we explain these things. The image file I work on in LR is generally a 16 bit file, and the monitor is irrelevant to the file and to Adobe. However, what I actually see is dependent on the properties either of my monitor or of my printer, or, if I post to the web, the iPhone, iPad or PC monitor that someone else is using. In each case a profile is needed to "translate" the image file to the device, and explain how out-of-gamut colours are dealt with: a monitor calibration profile, or a printer profile, coupled with a rendering intent. Because I have no idea what devices web users have, I export stuff to web in the lowest common denominator: 8bit sRGB.

Back in the privacy of my own home, I keep my image files, and process them, using the maximum amount of information available: the original pixel size, 16bit colour depth and Adobe RGB (the latter being the colour space used by my camera). (Edit: perhaps I should switch to ProPhotoRGB to allow more flexible post processing.)

So: colour space, bit depth, display device properties, printer properties - they're all different, although we can make them interact in many different ways.

[ajohnw]

I used a bad example in some ways. An 8 bit count goes to 255 and then rolls over to zero. A 12 bit count goes to 4095 and then roles over to zero. 16 x 255 is 4080 so there are 15 more fractional counts available so could be restored providing the 12 bit count doesn't overflow.

There doesn't seem to be enough technical information available on what this means in practice other than some packages I use retain data in a 32 bit floating point format. That is equivalent to roughly 24 bits of precision or 7 decimal digits worth. It's the same sort of format a n.nn.etc x 10^n. I'd guess that the math cores in PC use IEEE single precision 32 bit floating point but it's not possible to know just how the software uses this without looking. They could for instance assume numbers are scaled by some power so that effectively the decimal or in this case binary point is moved to the right. The whole idea of the extra precision is to avoid loosing information or generating excessive rounding errors as the numbers are manipulated.

Adobe I understand and some other packages use ProPhoto or do they? This implies that the numbers are translated to some sort of ordinates that correspond to the ProPhoto colour space based on some imaginary colour. On the other hand maybe they just dump the numbers from raw in them as it's a 16bit space so will leave head room. Pass I don't know. There have been comments in the past that Adobe actually use straight binary 16bit arithmetic which suggest they just dump the numbers in that and do with them what ever. Worse still it has also been suggested that the use signed 16 bit which mean in practice that they have 15 positive bits of precision. I doubt that because there are standard 3x16bit colour interchange format that are in use and I'm sure Adobe can handle them. There may even be bigger ones. All I know on that score is that an acquaintance who has work published gets asked where the rest of the file is when he hands over results from Canon FF cameras. The people who take them expect what medium format put out and have to convert them.

I do wonder when package offer a workspace setting of ProPhoto if they just mean a 3x16bit colour space and not something that has been translated to numbers in a true prophoto gamut. If they do translate I would strongly suspect that they retain them in floating point format to prevent errors when they are converted to other gamuts.

I did find a web page on one of the standards organisation that talked about low level colour formats and from memory it did not rely on associating this with a particular gamut. At the time I wasn't terribly interested, looking for something else but did look as info on standards sites is usually accurate - if anyone is using them. As I see it this level doesn't really matter. We adjust according to how a monitor is set up as that is how we see the colours we produce. The monitor might be set to sRGB, aRGB or even some paper simulation. The camera ICC file does the basic translation scaling colours according to it's sensor and it's filter performance. The next issue is if any data this produces is thrown away and when.

John
-

[Manfred M]

Adobe I understand and some other packages use ProPhoto or do they?

In Lightroom 5; you can select sRGB, AdobeRGB and ProPhoto; with the system defaulting to ProPhoto. Choice of 8 or 16 bit

Photoshop (ACR) has a wide range of operating profiles, including these three.