We helped to popularize the term, but it seems to cause some confusion.
The term ISO invariance is being used increasingly widely in camera technology circles, but the comments we’ve seen suggest it’s often misused and misunderstood.
This is partly our fault: ISO invariance is the term we chose to describe a phenomenon that was already being discussed, and our usage and tests have helped popularize the term. So we thought it made sense to make clear what it is, what it isn’t, and why it matters.
Sensor performance, not camera behavior
ISO invariance is an assessment of how much or little improvement in noise you get by increasing gain.
It’s an assessment of a sensor‘s performance, not a description of how a camera delivers its ISO settings.
Why add gain?
The best way to understand it is probably to take a look at why gain / amplification is used in cameras in the first place. If you take a photo in low light, your sensor has less light to work with and will generate a smaller signal from your exposure. Amplifying this signal before it’s digitized has two potential benefits:
- It helps match the signal voltage to voltage range that your analog to digital converter (ADC) is designed to digitize. (In other words, you make sure that the signal is captured using lots of your available Raw values)
- It reduces the impact of any noise that might be introduced after amplification (if you double the scale of the signal, then any noise introduced later has half the impact).
It also has a major cost:
- If you take the signal coming off the sensor and amplify it 2x (ie: add one stop of gain), then the brightest stop of highlights will be represented by a voltage that’s too great for your ADC, and those values will be recorded as its maximum value: they’ll be clipped.
In the early days of digital cameras, cameras would add more amplification every time you increased the ISO setting. This link between ISO and gain is no longer as rigid as it was, but in very broad terms, it’s still usually the case that most cameras typically apply more gain when you increase the ISO setting. How’s that for definitive?
Modern sensors are excellent
However, in recent years, sensor and electronics designs have improved to the point that there isn’t much post-amplification electronic noise to overcome. And, after a certain point, it’s worth questioning how much of a shadow improvement am I getting in return for the highlight information I’m throwing away?
And that’s what ISO invariance assesses: are there circumstances in which it would be better to keep the ISO down, and save the highlights, if there’s not much improvement in noise performance?
We opted to call this ISO invariance because the existing term “ISO less” sounded like an binary: something sensors either were or weren’t. We hoped the term “invariance” would make it clearer that sensors would have a degree of variance or, as it turns out, ranges over which they are less or more ISO variant.
In hindsight, ISO invariance isn’t the best possible term for this phenomenon, not least because it’s an assessment of the impact of analog gain, and ISO doesn’t necessarily depend on analog gain. But also because it got interpreted by some people as a binary position anyway.
ISO invariance does not tell you how a camera delivers its ISO settings
Another common misunderstanding of the term is that it means the camera doesn’t change gain when you change ISO. This is not the case
Delivering multiple ISO from the same gain level would be a method of exploiting a sensor’s high degree of ISO invariance, but it’s extremely unusual for a camera to do this. There are lots of cameras that deliver some
of their ISO settings without changing gain, or have modes that change ISO value without changing gain, but the majority of cameras do change gain when you change the ISO setting.
It doesn’t mean you don’t have to worry about what ISO setting you use
ISO invariance also doesn’t mean it no longer matters which ISO setting you use on your camera. If you’re in a mode with any degree of automation (Program, Shutter Priority or Aperture Priority), your ISO setting will change the exposure your camera tries to achieve. Setting ISO higher than you need will result in capturing less light than you might others, and this reduction in light leaves the image looking noisier.
But also, because most cameras increase gain as you increase ISO, some amount of highlight information is pushed to clipping unnecessarily and this can’t be recovered, later.
Why don’t cameras exploit this?
So far, we’ve not encountered a stills camera that exploits ISO invariance performance by separating exposure and image lightness from gain. There are modes that take some advantage of it, though: many cameras’ DR modes work by intentionally giving lower-than-normal exposure to a given gain level so that the camera captures and retains an extra stop or two of highlight information, then apply a different tone curve that accommodates these extra highlights.
But generally, though, exploiting an ISO invariant sensor is a bit messy: you have to set an appropriate exposure in a low light scene and then lower your ISO setting without changing this exposure, to prevent the highlights in that scene from being amplified into clipping. This usually results in preview that’s multiple stops darker than you want the final image to be. It’s a pretty clumsy, Raw-only way of working.
There is hope, though
Interestingly, it’s becoming more common on the video side of things. Sony’s FX series cameras (many of which share their core hardware with Alpha models), offer an ‘Exposure Index’ mode, where the camera always operates at the low or high conversion gain states of its sensor, then records ‘ISO’ as a metadata tag that’s interpreted when editing. Increasing the ISO setting doen’t increase gain, so you don’t lose highlight data.
Similarly, the biggest distinction between NRaw and R3D raw video modes on the Nikon ZR is the way ISO is handled. NRaw works like stills mode: adding gain each time you increase the ISO setting, but in R3D mode it sticks to the two native conversion gain states of its sensor. This means that you don’t push data to clipping when you increase ISO, you just change a metadata tag indicating how light or dark the footage should be rendered. This is what makes it possible to adjust the apparent “ISO” after the fact in editing software.
But, for now, we’ve seen no such mode in a stills camera. In part this is because it would require a fundamental rethinking of how exposure, gain and image lightness should be handled. And, for now at least, the industry seems happier to tinker with the increasingly strained-looking ‘it’s just like film speed’ metaphor that is ISO, than to think about how best to make use of modern sensor performance.
Want to stay up to date on the latest product news and releases? Add DPReview as a preferred source to ensure our independent journalism makes it to the top of your Google search results.
