# Thread: bit depth and models space

1. ## bit depth and models space

Hello, my name is Daniel and I'm new in Digital Photography.

I'm trying to understand the bit depth per channel and models space.

I know a colour model describes the colours you can see and work in a digital image. Some examples are RGB, CMYK, HSB...

A model space is a variant of a colour model. It has a range of colours. Some examples are sRGB, Adobe RGB, ProPhoto RGB...

Having in mind the spectral colour, ProPhoto RGB can represent more colours than Adobe RGB, and Adobe RGB can represent more ones than sRGB.

On the other hand, taking the sRGB model space as an example, it uses three channels (Red, Green, Blue) to represent a colour. Each channel can have 8 bits, so in this case we will have an image of 24 bits (8 bits x 3 channels) and it can reproduce 16,7 millions of colours.

But we can have images of 48 bits (16 bits per channel) that can reproduce more colours than images of 24 bits.

I suppose Adobe RGB represents colours in the same way (Red, Green, Blue channels) so, which is the reason Adobe RGB (or ProPhoto RGB) represents more colours in the spectral colour than sRGB if, for example, I compare an image with sRGB of 48 bits and another image with Adobe RGB of 24 bits?

I appreciate any help, thank you,
Daniel

2. ## Re: bit depth and models space

Hi Daniel,

Probably the easiest way to visualise what you're asking is to think of a staircase analogy; bit depth equates to the number of steps, but doesn't have any bearing on the height of the staircase (because we don't know the vertical interval between each step).

In a large colourspace - like Prophoto - the distance between the steps is bigger than the steps between levels in a small colourspace like sRGB. In practice, it means that the numbers used to describe colours are different, so if a set of numbers describes an image in, say, Adobe RGB colourspace -- and you get that image printed by a device that doesn't understand colourspaces and ASSUMES that the numbers are for an sRGB image (a common assumption) then the image that gets printed will be darker and less saturated.

Does this help?

3. ## Re: bit depth and models space

Maxwell's House podcast from the TWiT network http://twit.tv/mh numbers 56,57,58, and 60 contain a detailed discussion of color spaces and how and why to use them. I would suggest listening to several of these episodes.

-Sonic

4. ## Re: bit depth and models space

Start with a bit depth of 8. A pixel with Red = 100, Green = 150 and Blue = 25 will be a different color in Adobe RGB than in sRGB. Both color spaces have the same number of colors (8X8X8) but the colors will be different. The sRGB space includes only colors printable on common printers and monitors when the standard was established. The Adobe color space includes additional colors printable with modern printers using 8 of more colors and viewable on modern monitors. All of this has nothing to do with 8 or 16 bit depth.

Bit depth has to do with how much post processing the image can take. Jpeg images have 8 bits per channel. RAW images are stored in photoshop as 16 bit images but typically contain 12-14 bits from the camera. If you take a picture of someone with a red shirt on in the middle of very dense fog the shirt will not look red. The histogram of this picture will have all of the red color data lumped in the middle. If the picture had an 8 bit depth then the red level may only range from 125 to 135 which looks grey in the fog. You can stretch that out with a levels adjustment so that the red goes from 0 to 255 but you will still only have 10 different levels. This appears as banding in the image. If you do the same thing with a 16 bit image your output will go from 0 to 255 but there will be many more levels and hopefully no banding.

The bit depth has a large practical effect on post processing and not a meaningfull effect on color space when the image is not post processed.

Alan

5. ## Re: bit depth and models space

Colin, Sonic and Alan, thank you for you help.

The colour space is used to determine the range (gamut) of colours you will use, if you choose Adobe RGB you will be able to represent more colours than if you choose sRGB and the bit depth is used to describe discrete values between the range (colour space) you have chosen.

You choose a colour space among others depending on your purpose, if you want to display it on Internet using sRGB wil be enough but if you want to edit it maybe it will be better to choose Adobe RGB.

Finally having a discrete value (bit depth) in a colour space you can convert it into another colour space using coordinate systems. I suppose if you have a discrete value in a large colour space, the colour that this value represents will be approximated in a shorter colour space, I mean, the discrete value is the same but with a similar colour.

Your help was very useful to understand my question.

I have also posted my question here http://forums.dpreview.com/forums/re...hread=37791084. I referenced it to give more information about my doubt if someone has the same question.

6. ## Re: bit depth and models space

Originally Posted by user Daniel
You choose a colour space among others depending on your purpose, if you want to display it on Internet using sRGB wil be enough but if you want to edit it maybe it will be better to choose Adobe RGB.
Monitors capable of displaying colours outside of the sRGB gamut are starting to appear, but it's possibly best to think of sRGB as being "Safe RGB".

Wider colour spaces have their uses, but the bigger the space, the more care one needs to exercise; As a rule of thumb, if you're sticking to "photo realism" where you're trying to accurately reproduce captured colours then Adobe RGB is usually your best bet (although Prophoto is unlikely to do any harm) because the camera can capture and many printers can print (excluding "print shops") well into the Adobe RGB gamut; one may just be limited in ones ability to display some colours, depending on monitor.

The bigger problems occur when one shifts away from photorealism and starts manipulating image colours; at this point one can get into a situation where what's displayed on an sRGB screen is NOT the actual colour selected - and of course it prints a different colour (essentially just an out of gamut issue).

In summary,

* If one has a cheap monitor or has prints done "in town", stick to sRGB.

* If they do their own printing and have a good grasp of colour management, it's safe to use Adobe RGB or Prophoto colourspaces.

7. ## Re: bit depth and models space

I have learnt a little more and I have thought some situations. First what I have learnt:

A Colour Space defines the gamut of colours.

A Bit Depth defines the amount of colours, no matter which is the colour space, just the amount.

If you have 16 millions of colours (24 bpp) you can just represent 16 millions of colours in any colour space (of course, the gamut of these colours would differ) As Colin said, in his first example, in a large colour space, the distance between the steps is bigger than the distance of them in a small colour space. I mean, 6 steps (bit depth) will be larger in a large colour space than 6 steps in a small one. Correct me if I'm wrong.

Now, the two situations I thought:

Situation 1: you have an image of 24 bpp in sRGB and you convert it in an image of 24 bpp in Adobe RGB. You won't have problems because the gamut of sRGB is contained into Adobe RGB (they would be distribute separately and they would represent exactly the same gamut in sRGB). And once the image was converted to Adobe RGB you could edit it and work with more gamut. But then if you convert it again in sRGB some colours could be lost because the gamut of Adobe RGB is not contained into Adobe RGB. Is this correct?

Situation 2: you have an image with 24 bpp (bit depth) in Adobe RGB (colour space).
The image has just green colours and uses just a small portion of gamut that can be represented in sRGB. You could convert the image to sRGB and you would not waste the bit depth. Is this correct? See Influence on bit depth distribution
here https://www.cambridgeincolour.com/tu...obeRGB1998.htm

Could someone tell me if these two situations and what I have learnt are correct?

8. ## Re: bit depth and models space

Originally Posted by user Daniel
Situation 1: you have an image of 24 bpp in sRGB and you convert it in an image of 24 bpp in Adobe RGB. You won't have problems because the gamut of sRGB is contained into Adobe RGB (they would be distribute separately and they would represent exactly the same gamut in sRGB). And once the image was converted to Adobe RGB you could edit it and work with more gamut. But then if you convert it again in sRGB some colours could be lost because the gamut of Adobe RGB is not contained into Adobe RGB. Is this correct?
Two things to consider here:

The first part about a colour in sRGB being available in Adobe RGB is right - is wrong - and is right again. It's right because as you say, the smaller is contained within the bigger - it's wrong "in a digital world" because due to rounding, it may not be possible to define an EXACT equivalent in Adobe RGB (just imagine two staircases - each with a different height - and each with the same number of steps. Put them side by side and you'd find that there wasn't a stair on the right at EXACTLY the same height above ground for each step on the left) - but in practical terms is right again anyway because we can't resolve a change that small anyway. However, having said the last bit, it IS possible to get visible damage by converting between image modes too often (although you have to try pretty hard).

The second part depends on the changes you've made. If you've moved some colours into the Adobe RGB only gamut then yes, these will be out of gamut when you convert back to sRGB.

Situation 2: you have an image with 24 bpp (bit depth) in Adobe RGB (colour space).
The image has just green colours and uses just a small portion of gamut that can be represented in sRGB. You could convert the image to sRGB and you would not waste the bit depth. Is this correct?
More or less, not withstanding any rounding errors.

9. ## Re: bit depth and models space

Thank you very much for your help Colin.

Another question, have you any idea about coordinate systems to represent colours within any given colour space? I mean, you can have a colour space and you can use any coordinate system with that colour space. Some coordinate systems are RGB, HSV, HSL.

I can't understand the following situation. You have an image with 24 bpp in sRGB. The COLOUR SPACE is sRGB, the BIT DEPTH is 24 bpp and which are the COORDINATE SYSTEM?

This is related with my question here http://forums.dpreview.com/forums/re...hread=37791084

10. ## Re: bit depth and models space

The coorcinates are stored in rgb always. The editing programs allow you to choose the color based on a different model and some programs even work in a different model, but the data is always stored in RGB values.

-Sonic

11. ## Re: bit depth and models space

Sonic, does it mean I always select the colour model (RGB, CMYK, HSV...) but I never select the coordinate system?

It is difficult for me to understand!

12. ## Re: bit depth and models space

This means that when the file is saved on the hard drive the coordinates are always RGB. The program you edit it in may or may not use other coordinate systems.

Colin C.I.E.L.A.B is not a coordinate system or a method of representing color spaces nor does PS use it by default. The purpose of C.I.E.L.A.B is to be color space independent, and therefore is used to map between different spaces in any competent program. PS uses it to do color space conversion and you can force editing in C.I.E.L.A.B if you want to, though the edits are still mapped as if it was RGB.

-Sonic

13. ## Re: bit depth and models space

Originally Posted by Sonic4Spuds
This means that when the file is saved on the hard drive the coordinates are always RGB. The program you edit it in may or may not use other coordinate systems.
Ok, I know what you mean with this.

On the other hand I don't understand the following:

I imagine an absolute colour space as an infinite container of gamuts where each particular colour space (sRGB, Adobe RGB...) resided in it and you can access the gamut of that particular colour space by any coordinate system (RGB, CMYK...)

You can choose a colour space and use any coordinate system with it.

You can convert between different colour spaces but remain with the same coordinate system although its reference points change.

And you can convert between different colour spaces and change the coordinate system too.

On the other hand, I have read this information in the photoshop help:

'A color model describes the colors we see and work with in digital images. Each color model, such as RGB, CMYK, or HSB, represents a different method (usually numeric) for describing color.' I think this is about Coordinate Systems. Do coordinate systems mean colour models?

'A color space is a variant of a color model and has a specific gamut (range) of colors. For example, within the RGB color model are a number of color spaces: Adobe RGB, sRGB, ProPhoto RGB, and so on.' This says that in RGB colour model we have Adobe RGB, sRGB, ProPhoto RGB..., but it does not mention that sRGB, Adobe RGB... can be referenced with CMYK, HSV (coordinate systems)...

Can you explain me this?

Sorry if I can't understand quickly.

14. ## Re: bit depth and models space

I'll have a try, it might get a bit mathematical... (don't worry, no formula)

There is an 'absolute' colour space, which covers the full range of colours humans can see, it's the Cie Lab system. As it has three independent (orthogonal) variables to describe a colour, it also is a coordinate system. It uses real values (real as in numbers with a decimal point) as coordinates, so has (in theory) infinite precision.

On a computer, we can only display a limited selection of all visible colours. That selection is called the gamut of a display. Thus, a gamut is a part of the full visible colour space (as represented by the Lab system), or a slice out of the cake. So no need to have a collection of gamuts.

Now to RGB vs CMYK vs. HSV vs. ...
First, it seems as if CMYK uses four independent coordinates. That is a false impression, the K (for black) would not be needed in an ideal CMY system, it's there to improve grey tones.

All our colour systems use therefore 3 independent variables, or coordinates. Each of these coordinate systems can describe the full colour space, they are just different ways of looking at the colours, each using 3 different orthogonal base vectors. This implies that each of these can be translated in the others through a simple mathematical transform. Also, there's an infinity of possible colour models, of which a few are used in practice (as they correspond to a usefull way of representing colours)

That's the theory at least . In practice, as we work with integers, rounding errors make that the conversions aren't perfect, so multiple conversions will introduce visible errors.

Now, when we combine the two parts above (gamut being the part of the full available colour space we use, and the colour model being the coordinate axes we use), we can create an infinity of colour models, some of which have become more or less standardised: sRGB, Adobe RGB, different CMYK systems. Each of these describes the part of the colour space covered (the gamut), and how to translate the Lab coordinates to the integer coordinates we want to use (RGB, HSV, CMY, ...)
(this is not a simple linear mapping, there's also gamma and such to consider, but the above is the basis)

Again, changing from one colour space to the other is in theory lossless when using real numbers; as we work with (finite precision) integer values, every change of colour space will introduce errors.

Do keep in mind that all those colour spaces are three-dimensional, so the nice figures you see showing a colour triangle are cuts through a 3D space to get a 2D figure (imagine cutting the point of a cube, the surface of the cut is a triangle)

Hope this helps

Remco

15. ## Re: bit depth and models space

Can I just say, in order to reassure any others who may be wondering if they're the only person who can't understand any of the above (whilst still retaining respect for those who do and are engaged in the discussion), that I got lost after....
Hello, my name is Daniel and I'm new in Digital Photography.
So, you're not alone.

16. ## Re: bit depth and model spaces

Let's me explain what I have understood.

We have an absolute colour space where reside the full range of colours that humans can see. We represent it using a coordinate system called Cie Lab that consists in three parameters (it is a 3D model):

a (green - magenta): negative values indicate green, positive values indicate magenta.
b (blue - yellow): negative values indicate blue, positive values indicate yellow.
l (lightness): a 0 value indicates black, a 100 value indicates white.

This coordinate system describes colours that humans cannot see too, use real numbers, is more accurate and is independent of devices. But there are more coordinate systems, some examples are RGB, CMYK, HSV, that let's us reference colours from the absolute colour space, but they are less accurate and use integer values. We can have infinite coordinate systems, they are just ways to represent colours, ways to access the absolute colour space although with less precision. The coordinate systems are called Colour Models too.

On the other hand, we have a gamut that is a part of the absolute colour space. A gamut is used by devices like monitors and printers. These devices can just represent a part of the colours we can see (gamut). We can have infinite gamuts.

When we combine a gamut with a coordinate system we get some standarized colour models, some examples are sRGB, Adobe RGB and ProPhoto RGB.

Finally, we can change between colour spaces using the same coordinate system, use the same colour space changing the coordinate systems and both, changing colour spaces and coordinate systems, the important thing is we lose precision changing between colour spaces.

I think I have cleared up my doubts! Thank you!

Daniel

17. ## Re: bit depth and model spaces

Originally Posted by user Daniel
Let's me explain what I have understood.
All looks good to me

In reality, each have their advantages and disadvantages (personally, I like working in LAB colour a lot) ... in the end, they're all just tools in a toolbox.

18. ## Re: bit depth and model spaces

The main thing is not that you loose precision when converting, but that some color spaces are better fore some thing than others (sRGB for displays csRGB for editing etc...) Any good converter will be lossless.

-Sonic

19. ## Re: bit depth and model spaces

Thank you Colin for your approval. Thank you Sonic for your correction.

I feel more comfortable and I'm excited to learn more about photography. Having the approval from people who know about photography is great to learn!

20. ## Re: bit depth and model spaces

I like when people post fun questions that give me a chance to research.

-
Sonic

Page 1 of 2 12 Last