Should an ISO Upper Limit be Set?

[drr1531]

In another thread discussion the ISO performance of the D500, I questioned the practice of using the upper ISO limit which many people seem to do. My own experience has been that underexposure(at any ISO setting) results in worse noise than properly exposing at higher ISO levels. So I asked, "what am I missing?"

In response, Nick(Palouse) provided the following:

... Raising ISO is really amplification of the signal that comes from the sensor. As ISO increases, and/or shadow areas increase, the amount of noise incorporated in the signal also increases. At some point this overwhelms the image ... Increases in ISO are done first with analog amplification and then at some point the analog amplifier runs out of steam...

Which is correct. Then:

...Shooting at a lower ISO than required for a "perfect" exposure will also protect highlights. "Exposure" can then be selectively adjusted in post, and with lower ISO you have greater Dynamic range...

Which is true once ISO levels beyond the analog amplification range are reached and in-camera digital "gain" is used to further boost signal(i.e.increase ISO). If raising exposure further has to be done digitally then in-camera or during post simply depends on which software does a better job.

OK, so point 1 is relatively clear. At ISO settings above the analog range it is really a choice of whether we think our PP software and skills do a better job than Nikon's in-camera digital signal boost. So personal preference.

However, when shooting at ISO settings within the analog range, which is better, higher ISO or under exposing and correcting in post? When you under-expose at a given ISO(aka gain) setting you're reducing the signal strength without changing the (read)noise threshold. Which means signal to noise ratio(S/N) gets worse. So which is worse, reducing signal or increasing ISO/gain? Historically my own experience has been that properly exposing regardless of ISO setting produces best results. But I'd never gone out of my way to find any data to support one argument or the other.

Well Bill Claff seems to have a lot of cred and his site, Photons to Photos, is packed with all sorts of technical information for those who care and insomnia cure for those who don't. But I've got one of those personalities and the curiosity was killing me so I spent some time there. And came across this article.

If one has the option to lower the ISO and the shutter speed (or widen the aperture), the highest S/N for the image is obtained by increasing the exposure, pushing the right end of the histogram right up to the upper edge of the range of exposure on the horizontal axis. This is the usual ETTR philosophy. Lowering the ISO one stop pushes the upper end of the dynamic range one stop to the right in absolute exposure, and pushing the histogram to the right climbs the rising S/N curve to better overall image quality.

Everybody get that? ETTR maximizes IQ
He goes on:

If on the other hand, one is limited by the subject matter (freezing motion, depth of field requirements, etc) to a given maximum EV, then it makes sense to raise the ISO to pull the top end of the camera's dynamic range down to the top end of the histogram; this has little benefit at that upper end, since all the curves are on top of one another in that regime. Nevertheless it improves image quality by raising the S/N ratio on the shadow end of the curves.

To summarize, adjust ss and f-stop until you run out of room then crank up ISO to keep the histogram pushed to the right.

Now here is where the upper ISO limit comes into play.

It ceases to make sense to raise the ISO beyond 1600 on the 1D3 in raw capture, because there is no improvement in read noise... It is better to underexpose at ISO 1600, since one obtains the same shadow detail with more highlight headroom...

Note this is a very old article that was based on testing with the Canon 1D3. The behavior is similar for most DSLRs but the top end of the analog ISO range varies by body, generally getting higher with newer generations. The analog limit for the 1D3 is actually ISO3200 but for that particular body there was very little to be gained with that last stop of ISO. Since then(2007) manufacturers have put a LOT of effort into improving low light performance so one can assume things have only gotten better since this work was done.

So, bottom line, increasing ISO up to the limit of the maximum analog gain setting produces less noise than holding ISO down and under exposing. So logically if one chooses to set an upper limit to ISO, the maximum analog setting would make the most sense. But what is that setting? There's no single answer. It varies by camera body.

I'm not sure how Bill Claff et al figure this out, but they have identified the points at which various camera bodies switch from analog gain/ISO to computational methods. Following are the max analog ISO settings for various Nikon bodies. Theoretically beyond these ISO values it is equivalent to raise exposure in camera or with PP.

D300 = 1600
D3 = 6400
D3S = 12,800
D7000= 1000
D7200= 8000
D4 = 25,600
D4S = 12,800(interesting...)
D810 = 2500 (this surprised me)
D750 = 8000
D500 = 51,200(wow!)
D850 = 25,600
D5 = 102,400

 

[Palouse]

Dan, a good summary and a great challenge.

I really like how this is worded:

To summarize, adjust ss and f-stop until you run out of room then crank up ISO to keep the histogram pushed to the right.

(underlining is my contribution).

ETTR makes theoretical and practical sense, to me, and in landscape work it is very easy to do--I'm not usually rushed for time as you wildlife folks are!

Like you, I wondered about the upper limits of ISO for normal shooting (shooting the once in a life time Bigfoot shot, and other such events are exceptions , crank it up)
Steve Perry provides this table that matches yours --except for a couple of cameras:

Subscribe to see EXIF info for this image (if available)

More to puzzle about. I think I'll contact him for an explanation, re D850 and D500 especially. My experience with the 810 was high ISO was not its strength.

This one I have to ruminate over a bit more: You wrote, "So, bottom line, increasing ISO up to the limit of the maximum analog gain setting produces less noise than holding ISO down and under exposing." I was of the opinion ( as stated in the other thread) that software could do a better and more selective job of pulling up shadows than the global approach that raised ISO would. I need to look at some of my recent shooting to be sure . But I do agree that the analog limit makes sense.

The other thing to consider is the nature of the picture. One with lots of dark areas would generate more noise; while those with unimportant highlights would not suffer from a bit of highlight loss. Different strategies required, right?

Anyway, thanks for adding to the discussion and bringing more logic and clarity to this.

 

[Pa]

These topics are discussed in Marc Levoy's lectures at a fairly deep theoretical level. I will need to review that section before attempting to summarize his discussion here. But I'm pretty sure he said to avoid going into the digital amplification regime.

 

[drr1531]

...Steve Perry provides this table that matches yours --except for a couple of cameras:...

It looks like Steve picked the ISO value where Claff starts to indicate "scaling"(which he does not explain) but still within the indicated analog portion of the curve. Presumably "scaling" in the analog region indicates some level of computational manipulation of the analog gain. From a practical standpoint the ISO values in question make it pretty much irrelevant for the vast majority of shooters.

 

[Ann_JS]

What remains clear to me that a bit of testing with our own cameras will be much more useful in enabling us to get the best possible performance from our equipment (and to find the limits to which we can push it when we need to!) than wasting much time on the theoretical charts produced by the various Measurebators.

Worth noticing is that the ISO 102,400 which is shown in some charts as "Maximum Analog ISO" for a D5 is clearly shown as the upper limit for ISO settings on the camera itself with the higher ISO speeds above that clearly stated to be "Extended" and delineated as a "High #" value.

There are also settings for "Low ISO" (below the Base 100 ISO) where some sort of de-amplification appears to be used.

I don't need charts to tell me what is clearly marked on the camera and which was clearly stated in Nikon's published specifications.

 

[drr1531]

What remains clear to me that a bit of testing with our own cameras will be much more useful in enabling us to get the best possible performance from our equipment (and to find the limits to which we can push it when we need to!) than wasting much time on the theoretical charts produced by the various Measurebators...

No question we should all run our own tests to understand the equipment that we work with. But measurebating has its uses for some of us. Sometimes it is helpful to validate one's subjective observations. Or to make one aware of confirmation bias/wishful thinking. Though some of us(many of us) make up our minds with very little and/or poor information and thereafter don't want to be confused by facts.

...I don't need charts to tell me what is clearly marked on the camera and which was clearly stated in Nikon's published specifications.

That is the case for the D5. Not all of the Nikon bodies are that way.

 

[Randy]

I only read the title and glanced at the rest....I'm not interested in how the sausage is made but

putting a limit on iso makes no sense as you are going to screw up something, SS or exposure or both, depending on what mode you are shooting...the whole point of auto iso is to let the camera select the iso needed for a perfect exposure (what the camera thinks is perfect)

 

[Mitchell]

A wise person once told me that "you can't fix blurry".

Sure I try to keep the ISO low and the histogram pushed to the right, but at the end of the day, you need enough shutterspeed at a given focal length to keep the photo sharp. These ISO charts are nice, but you can't fix blurry.