Understanding ISO: In digital cameras, it's not what you think it is

RedShark Replay: ISO ratings made perfect sense in the days of photochemical film but only serve to muddy the waters when it comes to assessing camera performance in the digital age.

I'm going to regret saying this, I know I am, but things were simple in the days of film – you read the number off the film can, dialled it into your meter, read the light, set the lens and turned over. Well, okay, that's possibly the most grotesque oversimplification in the history of detail-skimming understatement, but the point is that the sensitivity of photochemical negative was a straightforward specification that wasn't, usually, subject to much question. A director of photography keen on a grain-free image might overexpose half a stop or so, or might, if he or she was feeling arty, subtract one from the number on the meter before setting it on the lens (two if feeling very arty). In general, though, an ISO rating was an ISO rating, even if previously it had been an ASA rating, and push processing aside, that was that.

It shouldn't therefore surprise us to find that when electronic cameras started to move from the workaday reality of mere camerawork up to the heady realms of cinematography that the incumbents of feature film and high-end television production began asking what the native ISO of the camera was. It's an understandable question, albeit perhaps provoking a scowl from broadcast people who had spent a career growing comfortably used to the precision of a waveform monitor, which, after all, treats a video camera as an array of reflectance meters. Even beyond considerations of experience and familiarity, however, there are some fairly serious problems with even using ISO to characterise digital cameras, let alone establishing which number works best.

Chemicals, not digits

The principal technical difficulty is that the ISO exposure rating system, the most familiar part of which is closely based on the previous ASA system, was designed to characterise photochemical emulsion and not digital sensors. The calculation involves an analysis of the optical density achieved on a negative with respect to the amount of light which struck it, which is complicated by the fact that this response is highly nonlinear both in film and electronic acquisition, and in completely different ways (for film, it's referred to as the Hurter-Driffield curve). Most manufacturers specify the ISO of digital cameras – both still and motion picture – based on the settings which tend to create an image of equivalent apparent brightness to a notional photochemical picture, but since both technologies have such wildly differing responses to light, particularly in deep shadow and bright highlight, there can never be a precise equivalency.

This represents a fundamental problem to anyone attempting to simply meter for exposure with a digital sensor, especially in circumstances where the photographer's experience is mainly photochemical. The circumstance of a solely photochemically experienced director of photography being caught off guard by this situation is of course rare and growing rarer, but the fact that the industry as a whole continues to attempt to use ISO means that it's still a problem we have to deal with if we want to compare the native sensitivity of cameras.

Native sensitivity

But electronic cameras have traditionally had a native sensitivity, regardless of what we call it. They generally offer gain settings, often +3dB and +6dB (equal to a one-stop increase in sensitivity), which may look fairly OK, and higher numbers such as +9dB and +18dB which are typically useful only in emergencies. The technical implementation is identical – it's all just amplification – but with the implied existence of a baseline, standard sensitivity when 0dB of gain is selected. In many cases, this is why a published native sensitivity is clamoured-after in cameras which offer only ISO as a setting: because the cinematographer wants to know what the manufacturer considers normal, which is likely to be a sensible tradeoff between sensitivity and noise. Some cameras, such as Sony's FS700, permit a selection of either in the user interface, which, if nothing else, permits us to evaluate the company's assessment of the camera's native ISO.

And yes, it is still about the manufacturer's opinion, at least to some degree, even without the Recommended Exposure Index provisions in the most recent revision, ISO 12232:2006. Without naming names, it's been commonly understood for some time that one major camera manufacturer's ENG cameras are about half a stop more sensitive than another major manufacturer's, but also about 3dB noisier, implying that the underlying technology is of similar basic capability (3dB is almost exactly half a stop of gain).

Even this is complicated, though. Sony's A7S camera, which has been received with rapturous applause for its high dynamic range and sensitivity, is a case in point. Wide though the range is between full capacity white and unacceptable noise, the point of normal exposure is actually fairly far up the range, at least in the default picture styles. This means that there are fewer stops of dynamic range above the midtones than there are below. This stands in stark contrast to the capability of photochemical technology, where the exact opposite tends to be true. Expensive cameras such as Alexa are more capable of recovering grossly overexposed highlights in a way that's reminiscent of film, but this capability is still only approximately represented by an ISO figure.

A mixture of technical necessity and engineering opinion

There is an argument about how much the imbalance of highlight and shadow range on any camera is provoked by the manufacturer's decision to use a particular range of the sensor's output as midtone information in the final image. Much of it is a mixture of technical necessity and engineering opinion which can, in theory, be modified by loading different curves into the camera, or other techniques. Whether such an image might have a luminance response that would be pleasing to the eye is another matter, and the practical reality is of course that we can swap shadow detail for highlight detail at the cost of noisier shadows. Either way, experienced cinematographers know that if they're stuck with a camera with poor highlight handling, simply underexposing half to a full stop and grading back up in post production can have a beneficial effect.

So, this gets complicated, and trying to relate one's technical strategy in any particular situation to the ISO system is an exercise in frustration. We're planning more coverage of these sorts of techniques, including ways to make the best possible use of data space in both 8 and 10-bit systems as well as exposure concerns arising therefrom, in the near future – but we won't be relying solely on ISO when we do so.