ISO and noise - different points of view

[Tony Beach]

Here is a demonstration of the effect of lowering ISO on noise:

See, lowering ISO increased the noise in this instance. Ergo, raising ISO does not raise noise (at least in this example using my D800), so that means something else is responsible for that (see above).

Correct. And if photographers learn that "increasing ISO increases noise", they are going to get this wrong.

If the context is P or Av or Tv mode with a fixed EC control, though, then it is true that raising ISO raises noise, because both ISO exposure index and ISO setting increase together.

Again, ISO is about compensating for exposure. Understanding that is fundamental to understanding its role in exposure (i.e., it is the inverse of using Exposure Compensation on your camera's meter).

It is about various things, and if the camera is in an auto mode, the ISO number setting is about reducing exposure as well as about compensating for the reduction. In auto modes, it is not the inverse of the +/- control.

Set your camera to Manual Exposure Mode (and I can set my camera to Auto-ISO doing that) and you will see a direct correlation between raising exposure (and with Auto-ISO that would be effected by Exposure Compensation too) and lowering ISO (or vice versa).

So, if we want to be clear instead of making ourselves or other people look silly, we say a little more, which removes all room for confusion for any reader capable of understanding the distinctions.

"Raising ISO", by itself, without a context, can mean three different things: raising ISO exposure index but holding ISO setting fixed, raising ISO setting but holding ISO exposure index fixed, or raising the two yoked. Without this clarification, "Raising ISO" doesn't even mean anything specific.

What raising ISO doesn't mean is that it increases noise. You only arrive at that wrong conclusion by confusing it with being an integral part of exposure (it is an ancillary part of exposure, same as Exposure Compensation).

Yes. The difference being that the ISO number tells the processing software how much to brighten the picture, while the +/- control does not.

I tell my processing software what to do, not the other way around. Anyway, when I set my EC to expose hotter (ETTR) or not (ETTL) I end up with a file that has either less or more noise in it. OTOH, if I'm constrained in my exposure and need to compensate for that by raising my ISO that doesn't mean raising ISO increases the noise, being exposure constrained raises the noise and raising the ISO often decreases the noise.

 

[alanr0]

If you just explain the actual facts it ends up being much simpler:

"Reducing exposure causes an increase in the noise visible in the final image."

This is not only simpler, and thus easier to understand by your logic. It also has the benefit of being true, so it doesn't require even more explanations when the user switches the black box into M mode and the link between ISO and exposure is broken.

But that does not fit the facts of auto-exposure modes in which the exposure is not something that the user can set directly. So telling the user how the noise depends on the exposure is irrelevant and unnecessary if he cannot set both shutter speed and aperture.

If a complete beginner wants to know how to use P mode on his camera, then the simplest thing to tell him is that Increasing ISO will increase the noise. Of course, he will probably then ask why anyone would ever use anything other than the lowest ISO. Then you can go into more detail if he wants it, but he may be satisfied with knowing that he only needs to increase the ISO if the camera gives him a low-light warning when he is about to take the shot.

You post also suggests that you don't really understand the purpose of black-box models, which is to reduce the description of how the system behaves to its bare essentials. To describe how the outputs behave as mathematical functions of the inputs, pure and simple. To isolate the "what it does" from the "how it does it".

But that is exactly the point.

For a black box model to be useful, the predictions should be robust and reasonably accurate. Output image noise generally increases with ISO, but it matters by how much.

The appropriate mathematical function for a useful black-box model is camera-specific.

Suppose one is interested in deep shadow noise, where camera read noise dominates. Here is how 3 different cameras behave:

Loading…

www.photonstophotos.net

The Pentax K1 has low read noise which is near ISO-invariant, so that the output noise increases steadily with ISO setting.

The Canon 1DX appears to have significant analogue noise between pre-amp and ADC, so that there is a plateau between ISO 50 and ISO 1000 where read noise is almost independent of ISO setting. Here one can trade depth of field against shutter speed with minimal impact on output read noise.

The Sony 99M2 shows a step drop in noise at ISO 400, presumably where the DR pix gain switches.

Each of these calls for a different mathematical description. Arguably, only the Pentax shows a monotonic increase of noise with ISO setting, implied by your statements.

Above the deepest shadows, photon noise will dominate. Personally, in this regime I find it easier to think in terms of light captured than ISO setting but YMMV.

If you want to tell a beginner "reduce ISO to reduce noise", then this may work most of the time, but (depending on the camera) could easily take them a couple of stops in the wrong direction.

--
Alan Robinson

 

[Tom Axford]

For a black box model to be useful, the predictions should be robust and reasonably accurate. Output image noise generally increases with ISO, but it matters by how much.

The appropriate mathematical function for a useful black-box model is camera-specific.

That is an interesting question, but it is irrelevant to the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

Of course, I could have given a camera-specific function, but that would make no difference to the question, while introducing a red herring that would complicate the discussion.

 

[D Cox]

For a black box model to be useful, the predictions should be robust and reasonably accurate. Output image noise generally increases with ISO, but it matters by how much.

The appropriate mathematical function for a useful black-box model is camera-specific.

That is an interesting question, but it is irrelevant to the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

The answer to that question is that those people don't use any auto exposure modes.

Of course, I could have given a camera-specific function, but that would make no difference to the question, while introducing a red herring that would complicate the discussion.

 

[alanr0]

For a black box model to be useful, the predictions should be robust and reasonably accurate. Output image noise generally increases with ISO, but it matters by how much.

The appropriate mathematical function for a useful black-box model is camera-specific.

That is an interesting question, but it is irrelevant to the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

Given the statement in your original post "If the camera is in one of the main auto-exposure modes (Program, Aperture priority or Shutter priority), then increasing the ISO causes an increase in the noise visible in the final image (assuming all other camera controls are left unchanged)."

I imagine they take objection to it because, as presented, it is not strictly true in all cases.

In your follow-up thread, your alternative statements include a sprinkling of "usually" modifiers. While I would not claim these statements are misleading, I am reluctant to actively endorse them. This is more a case of "I wouldn't start from here", than factual disagreement.

Of course, I could have given a camera-specific function, but that would make no difference to the question, while introducing a red herring that would complicate the discussion.

That depends on the point you are trying to make, and where you are trying to steer the discourse. Your original unqualified statement was not strictly valid for some cameras.

You brought the concept of "black box model" into the discussion, and such models are indeed widely used and found useful within the engineering and scientific communities. While abstracting behaviour from detailed internal mechanisms is of value, benefits are contingent on the model accurately (if not precisely) reflecting real world performance.

As you stated "The behaviour of a system is described by treating the outputs as mathematical functions of the inputs". Understanding the specific mathematical function - at least in broad terms - is important if the model is to deliver useful predictions of performance.

If noise increases only slowly, or even decreases over part of the ISO range, that is something which a robust black box model should reflect.

 

[Tony Beach]

...the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

It's terrible advice because it leads to wrong conclusions. Should I set my camera to negative EC and boost the brightness in post? If I take your advice the answer would be yes, but in reality the answer is often no.

 

[Tony Beach]

For a black box model to be useful, the predictions should be robust and reasonably accurate. Output image noise generally increases with ISO, but it matters by how much.

The appropriate mathematical function for a useful black-box model is camera-specific.

That is an interesting question, but it is irrelevant to the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

The answer to that question is that those people don't use any auto exposure modes.

Wrong.

 

[robert1955]

For a black box model to be useful, the predictions should be robust and reasonably accurate. Output image noise generally increases with ISO, but it matters by how much.

The appropriate mathematical function for a useful black-box model is camera-specific.

That is an interesting question,

It is more than interesting, because if the answer is yes [and you sort of not disagree with that] your black box is faulty.

but it is irrelevant to the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

Now your changing tack: from trying to make a simplified teaching model, to wanting to discuss why some who have more knowledge than beginners disagree with the correctness of your simplification.

This sort of tacking BTW is an ever recurrent theme in ISO discussions: when

Of course, I could have given a camera-specific function, but that would make no difference to the question, while introducing a red herring that would complicate the discussion.

Sorry, but if there is a red herring it is the way you frame the objections to your simplification

 

[Iliah Borg]

...the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

It's terrible advice because it leads to wrong conclusions. Should I set my camera to negative EC and boost the brightness in post? If I take your advice the answer would be yes, but in reality the answer is often no.

Believe it or not, but I've seen more than few people who take "ISO noise" literally, and they do set ISO lower to avoid it.

 

[Roland Karlsson]

...the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

It's terrible advice because it leads to wrong conclusions. Should I set my camera to negative EC and boost the brightness in post? If I take your advice the answer would be yes, but in reality the answer is often no.

Believe it or not, but I've seen more than few people who take "ISO noise" literally, and they do set ISO lower to avoid it.

I think people here keep talking about different things.

The OP explicitly talked about changing ISO in auto exposure mode.

And - if you increase ISO then - then you decrease exposure, i.e. you get fewer photons into your system.

In that case - increasing ISO increases noise (relative to signal).

Of course - if you just change ISO and not change exposure - it is not clear cut what happens.

 

[Iliah Borg]

...the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

It's terrible advice because it leads to wrong conclusions. Should I set my camera to negative EC and boost the brightness in post? If I take your advice the answer would be yes, but in reality the answer is often no.

Believe it or not, but I've seen more than few people who take "ISO noise" literally, and they do set ISO lower to avoid it.

I think people here keep talking about different things.

The OP explicitly talked about changing ISO in auto exposure mode.

And - if you increase ISO then - then you decrease exposure, i.e. you get fewer photons into your system.

I increase ISO because I can't get away with a lower ISO because the exposure is low. Low exposure thus cause the noise, not ISO. Higher ISO helps to mitigate noise to some extent.

In that case - increasing ISO increases noise (relative to signal).

Of course - if you just change ISO and not change exposure - it is not clear cut what happens.

 

[Roland Karlsson]

I increase ISO because I can't get away with a lower ISO because the exposure is low. Low exposure thus cause the noise, not ISO. Higher ISO helps to mitigate noise to some extent.

Sure.

But that is not what the OP talks about. He explicitly specified auto-exposure.

The main problem here is that 96% of the posters are fully aware how everything works. Still we do not agree because we are talking about different things - and continue to do so until we reach 150 posts. Sad really

--
/Roland
Kalpanika X3F tools:

GitHub - Kalpanika/x3f: Tools for manipulating X3F files from Sigma cameras

Tools for manipulating X3F files from Sigma cameras - Kalpanika/x3f

github.com

 

[Tony Beach]

...the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

It's terrible advice because it leads to wrong conclusions. Should I set my camera to negative EC and boost the brightness in post? If I take your advice the answer would be yes, but in reality the answer is often no.

Believe it or not, but I've seen more than few people who take "ISO noise" literally, and they do set ISO lower to avoid it.

I think people here keep talking about different things.

Not me. I'm talking about a fundamentally flawed approach.

The OP explicitly talked about changing ISO in auto exposure mode.

Right, so what if I also change EC? Should I go positive or negative?

And - if you increase ISO then - then you decrease exposure, i.e. you get fewer photons into your system.

Again, do different than setting EC to a negative value.

In that case - increasing ISO increases noise (relative to signal).

That's just plain wrong (see above).

Of course - if you just change ISO and not change exposure - it is not clear cut what happens.

Very clear cut in this example, and this was done with one of the more ISO "invariant" cameras.

 

[beatboxa]

I increase ISO because I can't get away with a lower ISO because the exposure is low. Low exposure thus cause the noise, not ISO. Higher ISO helps to mitigate noise to some extent.

Sure.

But that is not what the OP talks about. He explicitly specified auto-exposure.

The main problem here is that 96% of the posters are fully aware how everything works. Still we do not agree because we are talking about different things - and continue to do so until we reach 150 posts. Sad really

--
/Roland
Kalpanika X3F tools:
https://github.com/kalpanika/x3f

As I explain in my post here , I think the problem is in a strange assumption: In what case would one shoot an auto-exposure mode (P, A, S) in order to minimize noise...?

A & S modes have 2 user inputs: ISO & one part of exposure.

A better technique when one has limited lighting is to shoot in manual + Auto-ISO, and to select the 2 inputs that are directly related to exposure: A & S.

If a user sets a maximum aperture that provides an adequate DoF; and the longest shutter speed that doesn't incur too much motion blur, then the user has minimized noise, while using the same number of inputs as an "auto-exposure" mode.

The auto-exposure modes help in the opposite scenario: where exposure / noise isn't the issue; but rather where highlight clipping is at risk.

In all cases that I can immediately think of, those auto-exposure modes should be used at or near base ISO; while manual + auto-ISO is the best 2-input route to minimize noise.

 

[Iliah Borg]

I increase ISO because I can't get away with a lower ISO because the exposure is low. Low exposure thus cause the noise, not ISO. Higher ISO helps to mitigate noise to some extent.

Sure.

But that is not what the OP talks about. He explicitly specified auto-exposure.

Auto-exposure is when one sets ISO, and the camera sets the exposure? So, if I set ISO too high, and camera decides on exposure that is too low, I caused noise, not ISO, and not the camera. For me not to cause excessive noise I need to understand how things work, and not be mislead.

 

[Pixel Pooper]

That is an interesting question, but it is irrelevant to the question raised in my original post, which was why some people take objection to the statement that increasing the ISO causes an increase in noise in auto-exposure modes.

The reason people object to that statement is because it feeds the myth that ISO causes noise and supports the concept of the exposure triangle. The beginners that get taught that way don’t stop there, they extrapolate to all kinds of incorrect conclusions, and then they go on forums and argue that “ISO noise” is caused by amplification, and that ISO is part of exposure. They write blog posts and create youtube videos where they expound their misconceptions. They create authoritative looking websites like Cambridge in Color, Digital Photography School, and Photography life and even publish books like Understanding Exposure by Bryan Peterson. All of these sources of misinformation feed the myths about ISO and exposure and make it hard for beginners to find accurate information, especially when there are so many sources that corroborate the myths. While done with the best intentions, this type of teaching has led to one of the most insidious campaigns of misinformation ever seen. That’s why.

 

[AceP]

The reason people object to that statement is because it feeds the myth that ISO causes noise and supports the concept of the exposure triangle. The beginners that get taught that way don’t stop there, they extrapolate to all kinds of incorrect conclusions, and then they go on forums and argue that “ISO noise” is caused by amplification, and that ISO is part of exposure. They write blog posts and create youtube videos where they expound their misconceptions. They create authoritative looking websites like Cambridge in Color, Digital Photography School, and Photography life and even publish books like Understanding Exposure by Bryan Peterson. All of these sources of misinformation feed the myths about ISO and exposure and make it hard for beginners to find accurate information, especially when there are so many sources that corroborate the myths. While done with the best intentions, this type of teaching has led to one of the most insidious campaigns of misinformation ever seen. That’s why.

Well said! And for some reason, I can picture this as a great speech, set to goosebump inducing music:

Alas, I think it may be more like this:

 

[Tony Beach]

While done with the best intentions, this type of teaching has led to one of the most insidious campaigns of misinformation ever seen.

Okay, but now tell us how you really feel about it.

 

[Pixel Pooper]

Definitely more like the second one.

 

[Pixel Pooper]

While done with the best intentions, this type of teaching has led to one of the most insidious campaigns of misinformation ever seen.

Okay, but now tell us how you really feel about it.

I would love to, but the forum software doesn't allow Australian forms of expression...