When comparing sensor dynamic range, what are those numbers based on?

Question

Dynamic range EV's are bandied about all the time and I get the feeling that a) they're not on the same 'scale' and b) they're misleading in what they're indicating so I'm hoping somebody can clarify.

Question of scale: MF body manufacturers often quote DR values in the 12-14 EV range while the numbers for 35mm bodies are in the 5-6 EV range, these obviously can't be the same 'scale' since DxO publishes that MF and 35mm bodies have similar values: (12-14 EV).

Question of wtf: So what exactly are those two different measurements measuring? Is this the indication of where you can stil find detail in the highest and lowest EV or where 'useful' data is? If I created a scene and measured the brightest EV at +6 and the lowest EV at -6 would I be able to discern detail in the entire photograph or would I only notice detail between +3 and -3?

EDIT: Also, for a camera with a DR of 12 vs a DR of 14, what exactly does that mean in real world terms?

Answer 1

The problem is dynamic range is subjective, seeing as the definition of dynamic range (at least in terms of sensors) is the difference between the brightest and darkest details the sensor can record.

The brightest value a sensor can record is easily found by looking at what point the sensor photosites become saturated and thus can't record any extra information. Dynamic range then ultimately comes down to what point all discernible detail is lost to noise.

The benchmarking site DXO-mark defines dynamic range as the difference between saturation of the photosites and the point at which the signal to noise ratio hits 1:1, that is where the signal and noise are equal. It's questionable whether any real detail is visible when the SNR is this bad, however it's a convenient figure to use and easy to measure. You can read about their definitions and test procedure here:

• http://www.dxomark.com/index.php/en/Learn-more/DxOMark-database/Measurements/Noise

DPreview also measure DR in a similar way by finding the saturation point and then darkening the image until the noise reaches a certain level, but despite devoting an entire page on the subject, they don't mention what noise figure they consider to be the limit of the dynamic range!

• http://www.dpreview.com/news/0011/00111608dynamicrange.asp

Given their DR scores are lower than DXO-mark I assume they are a little stricture and adopt a lower signal to noise threshold. As for the 5-6 EV DR stated for 35mm bodies, that figure will most likely be a qualitative assessment by photographers with a more conservative view on what is an acceptable level of detail. The marginal amount of shadow detail that is detectable by a computer program is unlikely to be categorized as "usable" by photographers. However when benchmarking many sensors you have to have a quantitative measure of at what light level detail is lost so the signal to noise ratio is used.

---

While we're on the subject of dynamic range it's worth pointing out that the [measured] dynamic range of a sensor in good light will be greater than the dynamic range in poor light. This is simply a result of the fact DR is determined by shadow noise, as noise increases DR decreases.

There are however multiple sources of noise, in good light noise in the shadows is mostly due to the electronics, whereas in poor light noise mostly originates from the discrete nature of light (so called photon noise). Small sensor compact cameras with good electronics will thus have a very respectable dynamic range in good light. It's only when light levels fall that the ability of large sensors to capture more photons that gives them an edge when it comes to DR.

Answer 2

Jrista said:

Now, if someone can come up with a consistent, reasonably accurate test that can provide real-world photographers with a real-world idea of how much DR they can actually eek out of their camera .

My posting is in answer to Jrista's question and I hope it will shed some light on Shizam's original question. These tests represent what you can actually achieve in real-world conditions with your camera. I did these tests at ISO 200 since that is the speed I normally work with.

You can do it yourself pretty accurately without spending any money. Instructions follow below the graph.

Here is the result of a test I did myself. I interpret this as meaning that, under practical conditions, with my camera, at ISO 200, I can usefully get a dynamic range of 8 EV. DxO give a result of 10 EV while DPReview give a result of 8.4 EV (all at ISO 200).

It is useful to look at noise levels as well, see this procedure for measuring noise levels.

The procedure in outline is this:

• use a uniform white wall (or similar surface) as a target.
• put your camera on a tripod.
• take a photo to get your starting exposure. All your photos must be taken in RAW mode. You should get a mid grey image.
• now set your camera to manual exposure.
• take a series of photos at decreasing EV steps until you have pure black, add one for good measure. You will probably need at least six photos.
• take a series of photos at increasing EV stops until you have pure white. Add one for good measure. You will probably need at least six photos.

Now that you have your photos you will analyse them as follows

• open the first one in Ufraw (or similar RAW converter)
• set exposure to default, input curve to linear, output curve to linear, choose no profile, choose manual white balance and adjust to neutral.
• Under live histogram you will see the average RGB levels for the entire image. Take the average of the RGB values and use that as the average brightness of the image.
• repeat for all photos in this set
• you will now have sets of relative EV values paired with average brighness values.
• plot them on a graph as I have done above.
• this will give an accurate indication of the dynamic range.

The main limitations on the accuracy of this procedure is the accuracy of your shutter speed and aperture. You will see this effect as small deviations from an ideal curve. For better resolution you can do the test at half stop intervals.

Take all photos in a short enough time that changes in ambient light have no effect.

The standard way of testing the dynamic range of your camera is to use the Stouffer step wedge.
You can see their price list here.