Color depth

[Arne 65]

I'm not an expert in this matter and a bit lost between various color depth information and what I actually see or process at what stage: 24.6 bits (per channel, I guess) at base ISO for the sensor of my Nikon D850, 16 bits (per channel) for the RAW files (which can be kept in TIFFs but will be compressed to 8 bits when exporting JPEGs), 8 bits per channel or 24 bits per pixel in the histograms of my editing software (Capture One Pro) and 10 bits per channel (8 bit + FRC) for my monitor (BenQ SW2700). Could anybody help me better understand?

[Manfred M]

These are different, yet related issues"

The DxOMark colour bit depth reflects the number of bits of data that are required to represent all the colours that the camera can discern at the base ISO. The sensor dynamic range is a similar measurement that shows the number of stops of light that the sensor can record. Both numbers are of academic interest as they reveal the "quality" of the sensor rather than meaningful data that impacts the images we produce with the camera. In both cases, neither actually reflect what the human visual system can actually resolve on any specific output device (generally our computer screens). Even high end computer screens tend to have a contrast ratio of around 1000:1, or 10-stops when viewed in a dark room. Turn up the lights and the number of individual shades drops. This is why professional retouchers generally work in a room that is dimly lit. Generally we tend to want the work area to be less than 70 cd/ square meter.

I find it a bit easier to explain the dynamic range constraints than the colour bit depth. The human visual system is most sensitive to mid-tone colours. We have more issues differentiating shadow detail and highlights, so while the sensor can record that wide a range is really meaningless when we look at these colours.

So far as I remember, the D850 A/D convertor (analogue to digital) is 14-bits, but our computers work at 16-bits, so a couple of zeros are added to the output data to make it compatible with the technology we use to view and edit the images. TIFF and PSD files can be set to either 8-bit or 16-bit. Most editing software (as well as the camera histogram) uses 8-bit JPEG image files to calculate the histogram values. This means that when working with raw data, there is a bit of "headroom" at either end of the histogram. If the software suggest the shadow detail is crushed or the highs are clipped, you actually still have a stop or so of data at either end.

Your computer screen has a native 8-bit resolution and what the FRC does is to rapidly pulse the lights (a technique sometimes called dithering) to emulate more colours to emulate 10-bit colours Your Benq screen has two colour modes; sRGB and AdobeRGB. The sRGB mode is capable of showing about 35% of the colours that humans can see while the Adobe RGB mode takes that to around 50%; generally the more vibrant blues and greens. The 10-bit mode allows you to make out significantly more colours than 8-bit, but... Any colours that cannot be displayed are referred to as "Out-of-Gamut" and the driver software converts those colours into ones one can see.

I hope that this helps clear things up a bit.

[dje]

24.6 bits (per channel, I guess) at base ISO for the sensor of my Nikon D850

I think you mean 26.4.

Typically the ADC puts out either 12bit or 14bit data into the raw file. This is not converted to 16 bit until it reaches the raw processor. These numbers (12 or 14) are just the number of bits used to express the analogue values coming from the sensor.

The Tonal Range and Color Sensitivity values produced by DXOMark represent the effective number of grey levels(Tonal Range) or Color values (Color Sensitivity) a camera can produce given the noise present in the signal (mainly shot noise). These values are just numbers but are expressed by the number of bits required to specify this number (log2 of the number). See here.

In other words, even though you have say 14 bit data, you can't use every one of the possible values it can represent due to noise rendering two adjacent levels in-distinguishable.

Dave

[xpatUSA]

Originally Posted by Arne 65 24.6 bits (per channel, I guess) at base ISO for the sensor of my Nikon D850

I think you mean 26.4.

Lost me there, gents.

I do realize that "decimal bits" can result from a calculation such as 'effective number of bits' for an ADC but, not owning a D850, I have no idea where that number comes from.

[Manfred M]

I have no idea where that number comes from.

DxOMark: https://www.dxomark.com/Cameras/Nikon/D850

[Arne 65]

Thanks much Manfred. A bit clearer, yes, but not completely, which is probably due to my lack of basic understanding in this area. What I still don't fully understand is for example the recommendation of this article (https://www.cambridgeincolour.com/tu...terization.htm) to work in 16 bit depth when the histogram is based on 8 bit depth. But that's the "bit of headroom" that you mentioned, I guess. Still, I thought of the difference in depth rather than breadth, but that's probably wrong. Also, to put it in simple terms without numbers, I'm trying to understand better what breath or depth of the real data I'm seeing when I'm editing based on a histogram that uses 8-bit JPEG files and is shown on my monitor that can show more than that. I hope I'm still making sense...

These are different, yet related issues"

The DxOMark colour bit depth reflects the number of bits of data that are required to represent all the colours that the camera can discern at the base ISO. The sensor dynamic range is a similar measurement that shows the number of stops of light that the sensor can record. Both numbers are of academic interest as they reveal the "quality" of the sensor rather than meaningful data that impacts the images we produce with the camera. In both cases, neither actually reflect what the human visual system can actually resolve on any specific output device (generally our computer screens). Even high end computer screens tend to have a contrast ratio of around 1000:1, or 10-stops when viewed in a dark room. Turn up the lights and the number of individual shades drops. This is why professional retouchers generally work in a room that is dimly lit. Generally we tend to want the work area to be less than 70 cd/ square meter.

I find it a bit easier to explain the dynamic range constraints than the colour bit depth. The human visual system is most sensitive to mid-tone colours. We have more issues differentiating shadow detail and highlights, so while the sensor can record that wide a range is really meaningless when we look at these colours.

So far as I remember, the D850 A/D convertor (analogue to digital) is 14-bits, but our computers work at 16-bits, so a couple of zeros are added to the output data to make it compatible with the technology we use to view and edit the images. TIFF and PSD files can be set to either 8-bit or 16-bit. Most editing software (as well as the camera histogram) uses 8-bit JPEG image files to calculate the histogram values. This means that when working with raw data, there is a bit of "headroom" at either end of the histogram. If the software suggest the shadow detail is crushed or the highs are clipped, you actually still have a stop or so of data at either end.

Your computer screen has a native 8-bit resolution and what the FRC does is to rapidly pulse the lights (a technique sometimes called dithering) to emulate more colours to emulate 10-bit colours Your Benq screen has two colour modes; sRGB and AdobeRGB. The sRGB mode is capable of showing about 35% of the colours that humans can see while the Adobe RGB mode takes that to around 50%; generally the more vibrant blues and greens. The 10-bit mode allows you to make out significantly more colours than 8-bit, but... Any colours that cannot be displayed are referred to as "Out-of-Gamut" and the driver software converts those colours into ones one can see.

I hope that this helps clear things up a bit.

[Arne 65]

I think you mean 26.4.

Typically the ADC puts out either 12bit or 14bit data into the raw file. This is not converted to 16 bit until it reaches the raw processor. These numbers (12 or 14) are just the number of bits used to express the analogue values coming from the sensor.

The Tonal Range and Color Sensitivity values produced by DXOMark represent the effective number of grey levels(Tonal Range) or Color values (Color Sensitivity) a camera can produce given the noise present in the signal (mainly shot noise). These values are just numbers but are expressed by the number of bits required to specify this number (log2 of the number). See here.

In other words, even though you have say 14 bit data, you can't use every one of the possible values it can represent due to noise rendering two adjacent levels in-distinguishable.

Dave

Thanks much Dave. I tried to read the article, but gave up, sorry. I'm a chemist and biochemist, not a physicist...:-).

[Arne 65]

Lost me there, gents.

I do realize that "decimal bits" can result from a calculation such as 'effective number of bits' for an ADC but, not owning a D850, I have no idea where that number comes from.

You lost me too, sorry. But no worries at all!

[xpatUSA]

Originally Posted by xpatUSA I have no idea where that number comes from.

DxOMark: https://www.dxomark.com/Cameras/Nikon/D850

Thanks Manfred (seriously)!

Good old DxO: A metric almost as bad as M-pix ...

"The best image quality metric that correlates with color depth is color sensitivity, which indicates to what degree of subtlety color nuances can be distinguished from one another ..."

[Arne 65]

Thanks Manfred (seriously)!

Good old DxO: A metric almost as bad as M-pix ...

"The best image quality metric that correlates with color depth is color sensitivity, which indicates to what degree of subtlety color nuances can be distinguished from one another ..."

So it's simply my eyes after all...

[xpatUSA]

You lost me too, sorry. But no worries at all!

Sorry about that.

Some measurements can be expressed in fractional numbers of bits even though, in computer-speak, a bit is either '0' or '1' and even though half-a-bit can not exist.

Such a measurement is "Effective number of bits" (ENOB) which has to do with the dynamic range of an analog-to-digital converter (ADC) as found in any digital camera.

https://en.wikipedia.org/wiki/Effective_number_of_bits

https://www.analog.com/en/analog-dia...-issue-90.html

My cameras have 12-bit ADC's but their ENOB is about 10.5 which means that each channel has only 1448 effective levels, not the popular '4096 levels' that so many assume.

Pardon my pedantry ...

[Arne 65]

Sorry about that.

Some measurements can be expressed in fractional numbers of bits even though, in computer-speak, a bit is either '0' or '1' and even though half-a-bit can not exist.

"Effective number of bits" (ENOB) has to do with the dynamic range of an analog-to-digital converter (ADC) as found in any digital camera.

https://en.wikipedia.org/wiki/Effective_number_of_bits

https://www.analog.com/en/analog-dia...-issue-90.html

My cameras have 12-bit ADC's but their ENOB is about 10.5 which means that each channel has only 1448 effective levels, not the popular '4096 levels' that many assume.

Pardon my pedantry ...

Not at all. Thanks much for the follow-up and further explanations. I actually practice pedantry as well, but in areas where I know a little more and feel more at ease.

[xpatUSA]

Not at all. Thanks much for the follow-up and further explanations.

You are most welcome.

I actually practice pedantry as well, but in areas where I know a little more and feel more at ease.

Good to hear. There's not many of us left ...

[dje]

I probably should have made it clearer that the terms "Tonal Range" and"Color Sensitivity" are terms specifically defined by DXOMark and not general terminology.

Nothing wrong with pedantry in moderation

I don't get too excited about some of DXOMark's metrics either, partly because it's hard to know exactly how they calculate them. I tend to take some of then as indicatiive only of some effect. In the case of the above two terms, it's just showing that noise has an effect on how useful the bit depth of the data can be. eg for most cameras, it would be a waste of time using a 16 bit ADC.

Dave

[davidedric]

I probably should have made it clearer that the terms "Tonal Range" and"Color Sensitivity" are terms specifically defined by DXOMark and not general terminology.

Nothing wrong with pedantry in moderation

There's EVERYTHING wrong with moderated pedantry

"Pedants of the world unite. You have nothing to lose except your sense of proportion."



Dave

[Arne 65]

I probably should have made it clearer that the terms "Tonal Range" and"Color Sensitivity" are terms specifically defined by DXOMark and not general terminology.

Nothing wrong with pedantry in moderation

I don't get too excited about some of DXOMark's metrics either, partly because it's hard to know exactly how they calculate them. I tend to take some of then as indicatiive only of some effect. In the case of the above two terms, it's just showing that noise has an effect on' how useful the bit depth of the data can be. eg for most cameras, it would be a waste of time using a 16 bit ADC.

Dave

Understood, thanks. I'll probably by Heinrich Kuhn's great quote. Thanks much for sharing.

[Arne 65]

Thanks for the further insights and the additional philosophical dimension...:-) Like everything in life, it's all a matter of balance.

[Arne 65]

You are most welcome.



Good to hear. There's not many of us left ...

Coming back to your earlier remark on M-PIX, you don't seem to be a fan. My experience with them was not great either. I have been searching for a good and yet affordable printing service for photo books. Any further thoughts?

[Manfred M]

I take a vastly different view of pedantry; and that involves both context and the audience.

In a group of "subject matter experts" a higher level of pedantry can be acceptable, especially if it makes a difference to the final outcome. If it does not, then we really have a negative situation that can be little more than "oneupmanship" where someone is trying to say "I'm more correct than you are!". In my view, that is always a negative. If I ask how much a car weighs and someone comes back with 1437.349012kg, it likely is irrelevant and the level of detail and the assumptions that go into that figure are probably more important.

When it comes to pedantry in dealing with non-specialists, it is usually not acceptable because it does not impart any meaningful or useful information to them. If the answer is not clear enough or more information is required, they will usually ask for it, without having them think they are not as knowledgeable as the writer.

As for the definitions that DxOMark uses, I will agree. Trying to quantify camera performance into a few single values is always going to be open to criticism. From a pedantic view, the critics are correct, but once again display disrespect to the target audience; people who want to know how "good" a camera is with respect to other similar (or even dissimilar) models. It seems that someone looking at a Hasselblad X1D-50c review is unlikely to be of interest to someone looking at a Panasonic Lumix DC-GX800. In fact giving a Sports (Low Light ISO) performance number for the Hasselblad seems to be rather amusing given the typical target audience for a medium format camera.

[xpatUSA]

Coming back to your earlier remark on M-PIX, you don't seem to be a fan. My experience with them was not great either. I have been searching for a good and yet affordable printing service for photo books. Any further thoughts?

Sorry, Arne, I can't help you there because I don't print.