ISO invariance

[Peter Connan]

Recently I have read about this term in a number of articles.

One guy goes so far as to say that astro-landscapes are best shot at relatively low ISO's (significantly under-exposed, as the shutter speed and aperture are capped by other factors), as the eventual noise will be effectively the same and Dynamic Range will be improved.

I did a little test, just two shots (one at ISO 800, the other at ISO8000, both 30-second exposures at f2.8 using a D750 and 15-30mm f2.8 Tamron. And while I can't seem to get the exposures matched perfectly, I can't really see much noise difference, although neither can I see much dynamic range difference.

However, after this I have been thinking about the implications: in theory, if you have an ISO-invariant camera, you need not be at all concerned about exposure, as long as you don't over-expose. In fact, you could even shoot everything in Manual mode, at your chosen shutter speed and aperture, and base ISO for maximum dynamic range, and just sort the exposure out afterwards.

The problem with this would be trying to figure out which photos are worth post-processing, as you would have to adjust exposure first...

In short, this flies in the face of all former training, and I would love to hear your views?

[Jakov Minić]

Peter it would be great to see your two test shots.

If it is true that you get similar quality results at different ISO levels (no surprise with a good camera like the D750), why not shoot at auto-ISO and get the exposure correct from the get-go?
With nowadays cameras I have no problems in shooting auto-ISO and prefer being able to control the aperture and shutter speed all the time.

[Akira]

I have read that using the lower ISO and underexpose is good to retain the highlight detail of highly contrasty images.  I'm not sure if that applies to the astro-landscapes.  All astro-landscapes I've seen have been shot at higher ISO.

[Matthew Currie]

Just for reference, though I doubt it will have much bearing on most who post here, a quick test shows that the Nikon D3200 is not ISO invariant.  The differences are not enormous at small steps, but they get significant at larger ones.  Here is a zoomed detail, too downsized for the serious pixel peeper, but it shows the difference between a shot at ISO 1600 and the same exposure at ISO 100,  upped by 4 stops in Capture NX-D.  As you can see, the exposure is a pretty good match, and the result is not overwhelmingly worse.  I can imagine that a stop or two of underexposure and subsequent opening of shadows, a sort of radical version of Active D-lighting, could get you ahead, but plain exposure compensation will not.

On the D3200, at least, if you're stuck for some reason with unchangeable aperture and shutter speed, Auto ISO would be the way to go.

[Tristin]

I under-expose by massive margins when I need the DR flexibility.  It's not always needed, but it can really do wonders.  Of course, this is with a D750.  The "Sunset and Trash" image I posted recently was pure black with a small sun in the middle at it's defult exposure.  Auto ISO is like anything auto; sometimes it works and sometimes it doesn't.  I still use it here and there but staying in manual and purposefully under-exposing when the scene DR gets wide is more reliable, given you shoot raw and PP.  I can always blow out highlights in PP if I'd like.

[simsurace]

Note that exposure is not influenced by ISO setting. ISO setting is the gain applied during or just before A/D conversion.
Here, we assume that aperture and shutter speed (and hence exposure) are determined by other factors and cannot be changed. The only thing that remains to be chosen is ISO, which simply determines by how much the electrical signals are amplified before being stored as digital values.

If you don't have anything in your image that might clip, getting the ISO right in camera is usually preferable, since you can then push the histogram to the right by bumping up ISO and use all the bit depth that is available to encode your pixel values. Intentionally leaving the histogram linger on the left opens the possibility of quantization noise (rounding errors). However, it has been argued previously that quantization noise is a rather minor source of noise (compared to read noise or photon noise, for example).

I think the main reason that always shooting at base ISO is not practical is that the camera is not equipped for that. You would need automatic normalization of the display in order to see what you shot (composition wise) instead of a black screen. The same goes with exposure-to-the-right; our cameras do not give us all the tools necessary for achieving truly optimal exposures. What we need to see is information on the RAW file itself. That information is currently only accessible on a computer using software such as RawDigger. You can get close by using a dedicated flat picture control and unitary white balance (UniWB), but your picture review will look awfully green, causing some discomfort. However, the truth is that getting within a stop or so of optimal exposure (erring on the low side because of the risk to clip something important) is in most cases sufficient.

I also think it is hard to make general recommendations for exposure. I prefer to determine exposure on a case-by-case basis with all considerations in mind with regards to final usage of the image. Any 'optimal setting' will be optimal only under certain specific assumptions/conditions.

[Bjørn Rørslett]

Any scene has a range of possible exposure values depending on how one wants the final image to appear. This combined with the dynamic range of modern cameras and the capabilities of the better conversion program now makes the concept of "wrong exposure" something fading away. Still, getting a decent span and distribution of the numerical data from the scene demands some skills not only from the camera but asks some efforts from the photographer as well. In this endeavour I see 'Auto iSO' as the equivalent of crutches for the infirm. They certainly help, but only on a temporary basis unless the illness condition is permanent. A tripod tends to be a more useful accessory.

[Ilkka Nissilä]

Cameras aren't really "ISO-invariant" as by increasing ISO you do get better image quality in the shadows than using a low ISO and adjusting the exposure correspondingly in post-processing. If you don't have a high scene dynamic range, you should increase ISO rather than let everything be underexposed. If you do have a high dynamic range scene then one option is to expose for the highlights and hope for the best in the shadows. However, capturing multiple exposures on tripod and combining the shots using exposure blending will lead to lower noise shadow areas in the final image. However, the use of multiple shots is not always practical, especially if the scene is changing. For scenes that are static enough, I do use the technique and am happy with what can be accomplished. I don't go overboard with it, and usually end up letting some shadows be black and highlights be slightly bright to create a natural appearance.

[Les Olson]

There are two sources of image noise: noise inherent in light, called photon or shot noise, and noise arising in the sensor.  (There is a good but dense technical discussion of the issues as they relate to CMOS sensors at http://micro.magnet.fsu.edu/primer/digitalimaging/cmosimagesensors.html).  When photons hit the silicon wells some are absorbed, and create a voltage, which is read as an analog signal and then digitised.  Before the voltage is digitised, it is amplified and the amplifier gain can be changed.  This is how changes in ISO in the camera's native ISO range are achieved - although ISO is not just the amplifier gain.  A second procedure can be used: multiplying the digitised values, and this is what the computer does when you move the "exposure" slider. 

There is a difference between these two ways of amplifying the signal: in the first case, noise introduced after the amplification - eg, in the A-D converter - is not amplified and the second case it is.  In some sensors this makes a difference to the result and in others it does not.  An "ISO invariant" sensor is one where it does not make any difference.  "ISO invariant" is a poor term, because the ISO is not invariant and has nothing to do with what is, but there it is.

There is no genuinely free lunch on offer, however.  You don't gain dynamic range, and you will still get posterisation. 

[simsurace]

"ISO invariant" is a poor term, because the ISO is not invariant and has nothing to do with what is, but there it is.

This potentially confusing usage of 'invariant' comes from Theoretical Physics, where it is commonplace ('a rotation-invariant system', 'a Lorentz-invariant action', 'a CP-invariant particle' etc.). According to this jargon, an 'X-invariant Y' is something of type Y that does not change when transformation X is applied to it. For sensors, the intended meaning of ISO-invariant is that the sensor (or, actually the whole camera) behaves the same no matter what ISO is set. Of course, this is only approximately true in some sensors, and false in most.

[elsa hoffmann]

Interesting thoughts

[Tristin]

There is no genuinely free lunch on offer, however.  You don't gain dynamic range, and you will still get posterisation.

I can shoot at iso 100, push it 5 stops and retain more data without impacting IQ and without having to combine brcketed frames.  I was sceptical about these caims myself until I tested it mysef.  I do not get posterization.  Technical under-pinnings do not change the fact that with the right cameras you indeed can get more DR with a little PP by heavily under-exposing.

[Peter Connan]

Thanks for all the interesting thoughts and information.

This article: http://www.clarkvision.com/articles/nightscapes/ claims that DR and colour representation is better when shooting at a lower ISO and boosting exposure in PP.

Here are two of the photos I took. As mentioned before, this is D750 + Tamron 15-30mm f2.8 at 30 seconds and f2.8. The one is at ISO800, the other at ISO8000.

The ISO8000 image is entirely un-edited, while the ISO800 image is lifted by 3.3 stops, but no other PP has been done.
I have also included two crops of each, one in a relatively bright area, the other in the darkest shadows. Unfortunately, the light on the foreground is from a campfire, and thus not constant.

I can personally see almost no difference in the sky (either in terms of noise or colours/dynamic range), but it does appear as if there is more detail in the shadow area of the lifted ISO800 image.

[Anthony]

I can shoot at iso 100, push it 5 stops and retain more data without impacting IQ and without having to combine brcketed frames.  I was sceptical about these caims myself until I tested it mysef.  I do not get posterization.  Technical under-pinnings do not change the fact that with the right cameras you indeed can get more DR with a little PP by heavily under-exposing.

I agree with this, at least in relation to the Fuji X-T1, which is said to be ISO invariant to a certain extent.  My technique for shooting interiors of churches is now to use the lowest reliable hand held shutter speed and the widest aperture which preserve highlights (e.g. stained glass windows) at base ISO (200), and to lift the dark regions in pp.  This gives me much better results than raising the ISO in order to have the main part of the church well exposed, because when I do that the stained glass windows are blown out (figuratively not literally!).  The results in the darker areas are equally good with either method, to my eyes at least.

Tripods are usually prohibited in such places, so combining exposures is usually not an option.

[elsa hoffmann]

the D800 clearly doesnt fall in the same category ..... in my opinion it sucks when lifting exposure in ACR. I little is fine - but 1 stop is too much.