dynamic range mistake?

[maxhodges]

Your article on dynamic range states

5-9 stops is generally all one can expect from the camera.

Are you serious? A dynamic range of 5 f-stops is equivalent to only 32 "zones". 6 f-stops equals only 64 zones. Are there many digital cameras on the market today that can only resolve less than 100 levels of tonal range? Can you name some?

Cheers,
Max Hodges
http://maxhodges.com

[dje]

Your article on dynamic range states

Are you serious? A dynamic range of 5 f-stops is equivalent to only 32 "zones". 6 f-stops equals only 64 zones. Are there many digital cameras on the market today that can only resolve less than 100 levels of tonal range? Can you name some?

Cheers,
Max Hodges
http://maxhodges.com

The article was probably written a few years ago and needs an update. Modern DSLR cameras seem to have a max dr between 11 and 14 stops at the base ISO. However if you are using high ISO eg 3200 you will lose 4-5 stops of dr.

Compact cameras would have lower dr.

Dave

[Manfred M]

The article was probably written a few years ago and needs an update. Modern DSLR cameras seem to have a max dr between 11 and 14 stops at the base ISO. However if you are using high ISO eg 3200 you will lose 4-5 stops of dr.

Compact cameras would have lower dr.

Dave

Not to mention reduced colour sensitivity and tonal range and more noise, as one increases the ISO.

[pnodrog]

Just checked on http://www.dxomark.com and saw a few compacts that at ISO 1600 were down at about a dynamic range 5 but most started at 10 or more at ISO 200. The technology is constantly improving so I guess any articles discussing it need to be updated at least every year.

Anyway Max welcome to C in C. I had a look at your web site and enjoyed the photojournalism style of the albums I viewed. I will go back and have a better look when I have more time.

[ajohnw]

This isn't precise but in principle a camera can resolve as many stops as the number of bits it can convert. Compacts often work to 10bits internally and produce 8bit images. DSLR's vary, say 12 to 14 bits. How many real bits as far as images are concerned varies. One way of getting some idea of that is to look at camera jpg curves in the reviews on Dpreview. This can give a good idea what the manufacturers manage to use. These are shown as graphs with 0-256 on one axis and stops on the other.

Screens can be a considerable limitation as well. Getting a true 8bit out of those can be a bit of a problem.

John
-

[William W]

. . . Are you serious? A dynamic range of 5 f-stops is equivalent to only 32 "zones". 6 f-stops equals only 64 zones. Are there many digital cameras on the market today that can only resolve less than 100 levels of tonal range? Can you name some?

Maybe not on the new market - but many folk still use older cameras.

WW

[Colin Southern]

Are there many digital cameras on the market today that can only resolve less than 100 levels of tonal range?

The dynamic range of the camera is primarily driven by the sensor. The number of levels of tonal range is primarily determined by the resolution of the analog-to-digital converter. If the camera has a typical resolution of 14 bits then it'll in-essence deliver 16384 tonal levels regardless of the dynamic range. Or to use the old staircase analogy; dynamic range is represented by the height of the staircase whereas tonal levels are represented by the number of steps.

[maxhodges]

Modern DSLR cameras seem to have a max dr between 11 and 14 stops at the base ISO. However if you are using high ISO eg 3200 you will lose 4-5 stops of dr.

According to some sources, the dr drop-off of a 5D Mark III is only about 2 stops at 3200. You have to crank it up to around 256,000 before you crush the dr by 5 stops. But I suppose there may be lower-end consumer models which don't hold up that well.

[maxhodges]

Just checked on http://www.dxomark.com and saw a few compacts that at ISO 1600 were down at about a dynamic range 5 but most started at 10 or more at ISO 200.

Yeah, that sounds more like it. It seems a bit inauthentic to define the best one can expect from these cameras in terms of ISO of high ISO--That's a bit like defining the acceleration performance of a Porsche from a starting position of 160 KPH.

Thanks for the kind words.

[maxhodges]

This isn't precise but in principle a camera can resolve as many stops as the number of bits it can convert.

Yes, this kind of conclusion is warned against in the original article (at the bottom of the Bit Depth section). The number of bits only defines the theoretical maximum dynamic range--to quote: "high bit depth only helps minimize image posterization since total dynamic range is usually limited by noise levels". The bits just define the mathematical headroom, but the hardware may not be able to perform up to that theoretical maximum. Kind of like having a "64-bit audio recorder" with a signal-to-noise range of 385.32 dB (or exactly 9,223,372,036,854,775,807 integer values), but the quality of the mic pre-amps, cabling, and microphone by result in performance below this theoretical max.

To use an analogy, the fiber optic line here to my home in Japan is said to have a theoretical maximum data transfer rate of 100 Mbps, but transfer speed never gets very close to that limit in practical usage.

[dje]

According to some sources, the dr drop-off of a 5D Mark III is only about 2 stops at 3200. You have to crank it up to around 256,000 before you crush the dr by 5 stops. But I suppose there may be lower-end consumer models which don't hold up that well.

Yes those figures are what DXOMark reports for the 5D MkIII. ie 12 stops dr at ISO100 down to 10 stops at ISO3200. On the other hand, for the Nikon D800 it reports 14 stops at ISO100 down to about 10 stops at ISO3200. It depends on the sensor.

Dave

[Colin Southern]

Yes those figures are what DXOMark reports for the 5D MkIII. ie 12 stops dr at ISO100 down to 10 stops at ISO3200. On the other hand, for the Nikon D800 it reports 14 stops at ISO100 down to about 10 stops at ISO3200. It depends on the sensor.

Dave

Or is there something else going on?

In the case of the 5D3 - if we assume that the noise floor stays the same regardless of ISO - and at ISO 3200 only 1/32 of the amount of light is hitting the sensor (5 stops less) - so were working with 5 stops less range of light between the noise floor and what's captured - and yet the DR is only dropping 2 stops.

Doesn't appear to add up.

[dje]

Or is there something else going on?

In the case of the 5D3 - if we assume that the noise floor stays the same regardless of ISO - and at ISO 3200 only 1/32 of the amount of light is hitting the sensor (5 stops less) - so were working with 5 stops less range of light between the noise floor and what's captured - and yet the DR is only dropping 2 stops.

Doesn't appear to add up.

Colin if you look at the DXOMark curves for Dynamic Range, for the D800 there is basically a linear decrease in DR vs ISO setting. For the 5D MkIII (and most other Canon's), the curve has a knee in in at lower ISO. So DR doesn't really decrease much vs ISO until ISO gets to about 400 or 800 and then it becomes linear.

Why is this ? Well as far as I can see from other reading (see this link and this), this is because at low ISO, the most significant noise contribution is coming from after the ISO amp (eg in the A/D converter). In this area of operation, the ratio of the peak level out of the amp to the noise is therefore not reducing much as ISO amp gain is increased.

Make sense ?

Dave

[xpatUSA]

There is also the matter of Unity ISO Gain, below which it takes more than one electron to cause one ADU change and above which ISO amplification is just amplifying read noise. I've also read that some cameras really step up noise reduction above a certain level. So a simple linear model of noise versus ISO setting may not apply, IMHO.

[ajohnw]

Yes, this kind of conclusion is warned against in the original article (at the bottom of the Bit Depth section). The number of bits only defines the theoretical maximum dynamic range--to quote: "high bit depth only helps minimize image posterization since total dynamic range is usually limited by noise levels". The bits just define the mathematical headroom, but the hardware may not be able to perform up to that theoretical maximum. Kind of like having a "64-bit audio recorder" with a signal-to-noise range of 385.32 dB (or exactly 9,223,372,036,854,775,807 integer values), but the quality of the mic pre-amps, cabling, and microphone by result in performance below this theoretical max.

To use an analogy, the fiber optic line here to my home in Japan is said to have a theoretical maximum data transfer rate of 100 Mbps, but transfer speed never gets very close to that limit in practical usage.

To use an analogy, the fiber optic line here to my home in Japan is said to have a theoretical maximum data transfer rate of 100 Mbps, but transfer speed never gets very close to that limit in practical usage.

100Mbps is great but few servers will "ever" allow that rate out. What it means in practice is that your ISP can enlarge their server farm and send you lots of short bursts at that speed. This is what has happened on ADSL. They max out the bandwidth that is available but don't fill it as far as a user is concerned.

The same graphs I mentioned can also be used to get some idea of real noise levels as they disappear into noise at the dark end.

ISO gets complicated but to a certain extent it can be regarded as gain - gain magnifies the inherent noise as well so it's reasonable to expect less stops being available at higher iso setting.

Expanding on Colin's post, I didn't want to mention AtoD's. An 8bit set up is capable of showing 8 even brightness steps as far as the human eye is concerned. This is only an approximation. The way we see is a little more complicated. The actual levels in these bits for equal perceived brightness steps go 1,2,4,8,16,32,64,128 and just to be awkward 255 rather the 256 as 256 exceeds the max count of an 8bit set up. So there can be 255 tonal values. 10bits or more just keep doubling 512,1024 etc.

Personally I suspect as far as DSLR cameras go 12bits is the maximum what ever the bit depth of the A to D is and 1 or 2 bits of that may be lost to noise.. Large format does go further than this - at a cost.

My feeling is that monitors are the main limiting factor. I wonder how many peoples monitors match this simple set up.

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

Personally I am not fond of that particular way of setting mid grey and think this one is better. Off topic now but may be of interest



This one for the dark end



And this one for the light end



The usual grey scale is often easy



Never been sure about the usefulness of CMY and RGB but for completeness





I'm sitting in front of a Dell precalibrated monitor and can see problems with these and probably still will after calibrating it myself. The black end grey scale has some fundamental issues - ambient lighting for one and the characteristics of LCD type monitors for another but calibrators will part compensate for the latter by altering the levels.

As an aside these images are open source and should be on my machine some where if a Mod would like them to add to the monitor tutorial. I believe the current mid grey is aimed at none sRGB monitors.

John
-

[McQ]

A dynamic range of 5 f-stops is equivalent to only 32 "zones". 6 f-stops equals only 64 zones. Are there many digital cameras on the market today that can only resolve less than 100 levels of tonal range? Can you name some?

Hi Max, thanks for the feedback. The typical dynamic range of cameras has been updated to reflect cameras made within the past several years. That being said, the core point being made there is still more relevant than ever: we need to be careful about the false equivalence between digital levels and dynamic range. The number of tonal levels in a file just reflect the level of precision, and do not necessarily correspond with absolute changes in lightness. You could have a camera with a one stop dynamic range and 1000 tonal levels, for example, but could also have a camera with 10 stops of dynamic range and just 100 tonal levels. Dynamic range and levels quantify two totally different aspects of an image.