Dynamic Range

| 2 images

If you have read a review about a digital camera within the past few years, you have likely heard the term “Dynamic Range”. This term became a more well-known and fairly important term upon the release of Nikon’s D800, as that camera broke records with it’s dynamic range. Many of you may know that more dynamic range is better than less, and some of you may base your purchasing decisions partly on whether a camera has more dynamic range. Dynamic range has also become a controversial subject, particularly within fans of certain brands. I hope to clarify the topic of dynamic range here today, as it is actually a very simple concept.

Many photographers or would-be photographers don’t really know what dynamic range is, however, and some may end up purchasing equipment they do not need due to misunderstandings about what this little term actually represents. To keep it as simple as possible:

Dynamic Range is the ratio between the brightest tones a camera can record, and the darkest tones that become indistinguishable from electronic system noise.

That is dynamic range, also termed DR. It is a ratio, much like the signal to noise ratio (S/N, or SNR). The key difference between DR and SNR is that dynamic range represents the fixed signal range of the hardware, and SNR represents the actual signal range of an individual image. Dynamic range only accounts for electronic noise generated by the camera itself, while SNR must also account for noise within the light itself. Electronic noise is usually termed “read noise”, as it is noise added to the signal when the image is read off of the sensor and saved onto your memory card.

Throughout this article, I have used Adobe Lightroom 5 for processing.

What does dynamic range do?

Before I delve into how dynamic range is computed or where the read noise comes from, the more relevant question for most readers is probably: “What does having more dynamic range do for me?” That is an excellent question, and the answer is fairly strait forward. Note the two photos I have shared with this post. These photos are of my living room. It is in the front of my house, facing north, and as such it gets no direct natural sunlight. I took these photos last year when I rented a Sony A7r. Both photos are from the same location, one with the A7r the other with my 5D III, with the same lens (I rented a Canon EF-E mount adapter for the A7r).

This scene is a “high contrast” scene…it has very bright parts and very dark parts. Anyone who has tried to photograph a similar scene, or tried to photograph landscapes near sunset, or cityscapes during blue hour, etc. will know that photographing these kinds of scenes can be difficult. You may have to make a tradeoff…block up the shadows to get the highlights…or clip the highlights to get the shadows. Given the option, many will take two or three “bracketed” images, and combine them later on to create an HDR or “High Dynamic Range” image. In these situations, the technical issue is that the dynamic range of the scene often does not fit within the dynamic range of the sensor. In my living room, both photos were exposed to just barely clip the highlights from the bright, sunlit houses outside my front windows just a little to maximize my use of the sensor’s dynamic range in both cameras (in other words, both cameras were exposed to include a range of tones covering fully white “overexposed” pixels down through the deepest shadows.)

Strait out of the camera, these two photos looked like this:

Dynamic Range - Strait Out of the Camera

Dynamic Range – Strait Out of the Camera

Again, anyone who has tried to photograph something like this has seen photos like the one above. What many photographers may not understand is that the image above actually contains more than mostly-black pixels. The image above, which came from the Sony A7r, was processed to create the version in the gallery attached to this article. That amazing ability to bring detail out of the blackness of a photo is the value of dynamic range. The more dynamic range you have, the greater your ability to “lift the shadows” or “pull the shadows up” while maintaining detail in the darkest parts of the image will be.

What’s the difference between more or less dynamic range?

Those with an attentive eye will notice that in the two versions of my living room photo, one has better “shadow detail” than the other. This is because the two cameras I used do not have the same dynamic range. Dynamic range is measured in stops, a concept familiar to most photographers as the doubling or halving of an amount of light. In the case of DR, stops means a doubling or halving of the tonal range between the brightest tones (the whites) and the point at which the darkest tones get lost to read noise. If you increase DR by one stop, you double the range of discretely discernible tones that the camera is sensitive to. If you increase DR by two stops, you increase that range four-fold. If you increase DR by three stops, you increase that range eight-fold.

The Sony A7r has 13.1 stops of native DR¥. That isn’t quite as much as the D800, which has 13.4 stops, or the newer D810 which has a whopping 13.8 stops of native DR. In contrast to these cameras, all of which use Sony’s Exmor sensor, the Canon 5D III has 10.97 stops of DR. That is a difference, between the A7r and 5D III, of 2.13 stops of dynamic range, which means the A7r is capable of discerning 4.54x more discrete tones. Just on a numerical basis, it sounds like the A7r is quite a bit better.

So, what does that mean? What is the difference between having more or less DR? It means your images will be less noisy overall if you have more dynamic range. It means that gradients should be smoother and less noisy. It means that color should be more accurate, particularly in the midtones and darker. It means that the dark shadows will have more useful information that can be recovered during processing. A simpler way to put it is, your image quality will increase.

Increases in image quality with more dynamic range are always real, however there is a perceptual limit on how much the human eye can actually percieve such increases. This is often where the controversial nature of dynamic range comes into play. Many will advocate that having more than 11 stops of dynamic range is unnecessary, that they don’t ever need more than that, or that they can simply use HDR if they ever need more. These are entirely valid points. Not everyone photographs high contrast scenes that may actually be able to make use of two or three more stops of dynamic range. In brighter tones, those above a midtone 18% gray, our ability to perceive the differences in IQ drop considerably. Those with a keen eye may notice slightly less noise, particularly on smooth areas of an image such as a blue sky. Some may note slightly better color. It’s in the darker tones, those below a midtone 18% gray, that increases in dynamic range become more valuable, and in the shadows that they become particularly useful.

Examining the benefits

Let’s take a look at the deep shadows of the images above. In the case of both photographs, I simply started increasing the exposure by one stop at a time. Here is a sequence of images, from both the A7r and 5D III, with one to five stop exposure pushes in one stop increments. The A7r is on the left, the 5D III is on the right, starting with a 1-stop push:

Dynamic Range - One to Five Stops - Comparison

Dynamic Range – One to Five Stops Push – Comparison

In these small images it is difficult to see what’s going on in the deepest shadows, although you can generally see that by the fourth stop of pushing, the 5D III has started to break down in the shadows. You can even start to see banding on the walls by three stops. I’ve pulled out a crop from the back side of one of my recliners, where the deepest shadows exist, for stops three through five for a better comparison. Again A7r is on the left, 5D III is on the right, starting with a 3-stop push:

Dynamic Range - Crop - Three to Five Stops Push - Comparison

Dynamic Range – Crop – Three to Five Stops Push – Comparison

The breakdown of the 5D III as we continue to increase our push is much more obvious here. Even by three stops, the amount of noise in the 5D III image is higher, by four stops the characteristic banding of Canon sensors is revealed. (It should be noted that Canon has dealt with their banding issues to a degree with newer cameras. The issue has not been entirely resolved, but the 6D and 7D II have both demonstrated reduced banding, and the 5Ds should also exhibit similar improvements. Overall read noise, however, has not improved, so the underlying random noise level and the characteristic blotchy color has remained the same with newer Canon cameras (5Ds excepted, as it has yet to be tested.)) The A7r, in contrast, has not only held up well right through an insane five-stop push, it maintains good color quality, it still resolves detail in my carpet, wood grain is visible right down into the deepest shadows on the lower back side of the recliner.

Beyond the Extreme to Unreasonable

Before I delve into the fundamentals of dynamic range, and why it affects image quality in this way, I thought I would see how far I could push the A7r. With 13.1 stops of dynamic range, it held up extremely well in the deepest of shadows with a give stop push. Now, to be totally fair, few photographers will actually find a common, regular need to push their RAW file exposures around this much. Landscape photographers, interior photographers, maybe a few others, might find a need on a more frequent basis than most photographers, however in general, pushing more than three stops is fairly unusual.

That said, when you do need to push your exposures around by an extreme degree, having two more stops of dynamic range can be extremely useful. You may be able to avoid resorting to HDR, which increases the amount of work you have to do in post, and has it’s own set of issues in the face of extreme dynamic range (ghosting and posterization, an inability to deal with motion, etc.) So, just how far could you push an image with over 13 stops of dynamic range, assuming you apply some noise reduction to handle noise in the deepest shadows?

In these examples, I used a +5 stop exposure push, -100 highlights pull, and +50 shadows push. In addition, I increased sharpening to 65, and increased luminance NR to +40 w/ 75 detail on the A7r image, and +80 w/ 60 detail on the 5D III image. The 5D III image couldn’t handle more than 60 detail, it started to get artifacts, and even with 80 NR it still has heavy noise and low detail.

Dynamic Range - Extreme

Dynamic Range – Extreme

I think the results speak for themselves, so I’ll leave the practical explanation at that.

The Math

There are some fairly simple mathematics that are used to model dynamic range and signal-to-noise ratio. For those who are curious how dynamic range is calculated, here it is in a nutshell. First, dynamic range is a ratio, so the simplest math would be the following:

DynamicRangeRatio = MaximumSignal / ReadNoise

The maximum signal and read noise are generally derived, and deriving that is a more complex endeavor that I won’t go into here. For the purposes of this article, I’ll simply reference sensorgen.info for data. At ISO 100, the A7r has a maximum signal of 47081e-. The e- there denotes electrons. The A7r also has read noise of 5.3e-. Similarly, the 5D III has a maximum signal of 68151e- and read noise of 33.6e-. The dynamic range in N:1 notation for both cameras is 8883.21:1 (A7r) and 2028.3:1 (5DIII). By this simple ratio, you can clearly see that the A7r has more dynamic range.

This simple ratio is also a measure of the discrete number of tones that can be generated by each camera. This is an important concept, but also a potentially confusing one. In simplest terms, the higher the noise, the greater the difference between the brightness of one tone and the next must be in order to actually determine or perceive the difference. Both the A7r and the 5D III are 14-bit cameras, meaning their maximum potential tonality is 2^14:1, or 16384:1. It is clear that neither camera can make full use of the tonal range offered by storing 14-bit data. The 5D III, with it’s high noise levels, isn’t even able to make full use of 11 bits of information, which allows for 2048 discrete tones. The A7r is able to make full use of 13-bits of information, which allows for 8192 discrete tones…the excess of 691 tones requires the use of 14-bit data.

Dynamic range is rarely expressed as a simple ratio like this. It is most often expressed either in decibels or stops. Calculating what the dynamic range is in decibels requires some slightly more complex math involving a logarithm:

DynamicRangeDB = 20 * log(MaximumSignal/ReadNoise)

If we run our two cameras through this for ISO 100, we end up with 78.97dB for the A7r and 66.14dB for the 5D III. For most people, these values don’t mean anything, so it helps to convert them to stops, which is simply done by dividing by 6:

DynamicRangeStops = 20 * log(MaximumSignal/ReadNoise) / 6

That gives us 13.16 stops for the A7r, and 11.06 stops for the 5D III. That is a difference of ~2.14 stops of dynamic range. Denoting dynamic range in stops is useful, as most photographers are able to understand what it means to be able to increase exposure by a stop. The A7r is, thus, capable of recording shadow detail as if you had increased exposure by over two full stops on a camera with only 11 stops of dynamic range.

Dynamic Range vs. Signal-to-Noise Ratio

One last thing on dynamic range before we go. The math for dynamic range above is very similar to the math for signal-to-noise ratio. It is important to note that the two are not the same. Dynamic range (in this context) describes the capabilities of the hardware, whereas SNR describes the actual characteristics of an image. The maximum potential of the image is limited by the dynamic range of the hardware, however there are other factors that affect SNR that do not affect dynamic range.

The most important difference is that SNR requires a signal. That signal is your image, the one projected by a lens onto your sensor. Every signal has an intrinsic amount of noise, and that noise is on top of the noise of the electronics the image is being recorded by. While DR is the ratio from the brightest recordable tones in a sensor, and that sensor’s read noise, SNR must factor in the additional noise in the image signal itself. That gives us the following formula (in it’s most basic form):

SNR = Signal / SQRT(Signal + ReadNoise^2)

The first thing to notice here is that the noise of a signal is equal to the square root of that signal. If you have a signal strength of 10000e-, the noise in that signal is 100e-. The second thing to notice here is that the noise in the signal adds to the read noise, but it adds in quadrature. Read noise is usually already represented as an RMS, therefor it must be squared before being added to the signal and taking the square root in order to derive the total amount of noise in a signal.

Assuming we have an even darker medium-toned image where the median signal strength in each pixel is 10000e-, we can compute the SNR for both the A7r and 5D III. The A7r SNR is 99.86:1, whereas the 5D III SNR is 94.79:1. Not a huge difference…about a tenth of a stop. What about a signal of 1000e-, which would be a fairly dark shadow? The A7r SNR is 31.2:1, the 5D III SNR is 21.7:1. That is a difference of about half a stop. What about a signal of 100e-, which would be a very deep shadow? The A7r SNR is 8.8:1, the 5D III SNR is 2.85:1. That is a difference of 1 2/3rds stops. The A7r, in the deep shadows where there is very low signal, would be like exposing a 5D III one and two thirds stops longer. If you were to blend an HDR image with a 5D III, and wanted to get those really deep shadows, your dark exposure would need to be at least 1 2/3rds stops darker than the middle exposure, making the bright exposure 1 2/3rds stops brighter.

In mathematical terms, that is what a sensor with more dynamic range brings to the table. Higher signal to noise ratio in your actual images. You can expose, in a single shot, what it would take a camera with less dynamic range an HDR blend to achieve.

Canon Cameras and More Dynamic Range

When it comes to Canon cameras, with their high read noise, there is another comparison that can be made. The A7r, with a shadow signal of 100e- (which, BTW, is not even the darkest shadow signal the A7r could record…you could very well record a signal down to a little as maybe 12e- and still be able to get useful information out of it…whereas with the 5D III your utter limit is going to be closer to 70e-), is about 1 2/3rds stops better than the 5D III. As far as the shadows go, this is similar to exposing the 5D III 1 2/3rds stops longer…but it is also similar to exposing with 1 and 2/3rds higher ISO setting. There is a project called Magic Lantern that has produced some enhanced firmware for Canon cameras. Magic Lantern adds a lot of useful functionality to Canon cameras, including some dynamic range enhancement. For those willing to try, Magic Lantern can do exactly this…for a single photo, automatically read the sensor twice, once at ISO 100, once at a higher ISO like 1600, and combine the data from both reads in such a way that the shadows become less noisy, while the brighter tones are untouched. This can give Canon cameras over 13 stops of dynamic range. There are some additional steps to making your images usable, however if you need the increased dynamic range, it is a free option that could expand the life of your existing camera, saving you the need to buy something like a D810 (and likely a bunch of new lenses to go with it) or an A7r (and an adapter, at the very least, to use your existing Canon lenses).


¥ I use the term native to refer to the dynamic range measured by unmodified images that come out of the camera. There is also normalized DR, where images are reduced in size to a common baseline, which makes dynamic range among different cameras more comparable, however at the moment I do not have enough data to generate such normalized results for use in this article. Native DR is also what you get when working with RAW files regardless.