ISO and signal to noise ratios

[gollywop]

I really don't want to get into a back-and-forth here. This is not a topic that is going to be readily resolved and formed.

Maybe because phenomenological descriptions are too convoluted Reminds of the old Blind Men Appraising an Elephant fable.

It is much simpler IMHO to start with how ISO setting works. From there, anybody can deduce things like clipping,...

Yes. Understanding leads to enlightenment.

 

[Michael Fryd]

That is one amazingly thorough answer. Thank you!

The only time noise ever really bothers me is in deep shadows (say, the foreground in astrophotography), or in a blue sky. Astrophotographers usually recommend shooting at very high ISO levels and then to reduce the exposure in post, often with the explicit goal to increase the signal/noise ratio. Do you by any chance know what logic is involved in this reasoning?

Actually, it is not recommended to use "very high ISO levels." Rather, it is recommended to use the maximal exposure shooting conditions allow (at base ISO) and then, if ETTR is not possible (which is likely the case) and you are able to add ISO without clipping, apply ISO as high as needed up to, but not exceeding, the point where the camera becomes ISO-invariant. This ISO varies greatly for different cameras.*

Because light levels are so low in astrophotography, and the objects are moving (unless using timed trackers), maximal exposures (largest acceptable aperture and longest acceptable SS) are typically below ETTR. Under these conditions, it is not possible to achieve the full DR of the camera, so the fact that adding ISO reduces DR is not an issue. But the fact that read noise is reduced as ISO is added up to the point of ISO-invariance does increase the s/n and, hence, image quality.

Thus, the logic behind this is embodied in my earlier post.

On this issue, you will likely find this article of value:

http://dslr-astrophotography.com/iso-dslr-astrophotography/

* Once the acceptable aperture and SS have been determined for maximal exposure at base ISO, be careful that these settings are not altered when adding ISO. Under many circumstances when using programmed modes, changing ISO can also affect exposure settings. One is best off using manual mode so that the exposure settings (f-ratio and SS) are unaffected by the ISO setting.

 

[bclaff]

Michael Fryd wrote:
...

I talk about "visible image noise" because that;s the noise we see with our eyes. This avoids a discussion of absolute noise levels and Signal to Noise ratio.

...

You can avoid discussing Signal to Noise Ratio (SNR) but that is what determines whether noise is "visible".

Yes. But I wanted to leave that for a separate discussion. I don't want to bring in additional issues if i don't have to.

...

I'm sure it's a semantic fine-point.
Not bringing it into the discussion doesn't change the fact that you would be discussing "visible noise" without any foundation for making that objective rather than subjective.
If you want to start with a subjective conversation, that's you're business; I prefer to be more objective.

 

[FingerPainter]

I'm trying to come up with a short summary describing ISO on a digital camera. I'm not looking to get technical, or provide every detail. I just want to convey a short overview.

A worthy objective.

Would you say the following is reasonable, and how would you suggest improving it?

On most modern cameras ISO serves three purposes:

1. It sets the target exposure for the metering system - with higher ISO settings, the metering system will suggest lower exposures (less light)

I'd suggest that this isn't (or in good photography usually should't be) a purpose, but rather an effect. Other than those rare cases where one actually wants to increase noise why would one purposefully set a target exposure lower than it could be?

Furthermore, since a photographer should usually want to get the largest exposure possible within the constraints of DOF control, motion blur control and clipping, it seems to me that using ISO to restrict exposure is a much less optimal way to effect expsoure than to use the more direct exposure controls: aperture and shutter.

1. It may alter some of the operating parameters of the camera in order to optimize the camera's performance for the expected exposure

I'm not sure what you have in mind here, but again I suggest that this is an effect rather than a purpose. Why does the photographer care how the camera achieves a certain result?

1. It suggests a default mapping to use for converting the sensor data to brightness in the resulting RGB image

Why "default"?

I'd suggest that the purpose of the ISO setting is to set the image brightness level to be achieved from a given exposure level. In an autoexposure" mode, this brightness level is the target image brightness determined by the metering system as modified by the exposure compensation setting. Setting an ISO level can also have several effects, such as those you list, which might not be intentional on the part of the photographer.

I might add:

Generally the biggest factor in visible image noise is the amount of light on the sensor. As you increase the light on the sensor, the visible noise tends to go down.

High ISO setting are generally associated with high image noise because high ISO is generally associated with low exposures (less light on the sensor).

While many think of the high ISO as causing the noise, it is actually the low exposure. In fact, with many cameras, matching the ISO to the exposure will result in better quality than using a lower ISO and brightening the image n post processing.

I like all three of these paragraphs.

I talk about "visible image noise" because thats the noise we see with our eyes. This avoids a discussion of absolute noise levels and Signal to Noise ratio.

I do something similar. I talk about "noisiness" as opposed to "noise", because an image with a higher exposure actually has more shot noise, but a lower SNR and thus less visible effects of noise.

I've avoided the word "sensitivity" because I don't want to suggest that at high ISO, the sensor is "better" at detecting a small number of photons.

Yes. I think whenever one does include the word "sensitivity" in these discussions, one should preface it with "system output", but often it is best to just avoid the word entirely.

I have mentioned that an ISO setting that is typical for the expected exposure is frequently the way to maximize quality for that exposure.

I'm not sure about that.

Using lower ISO and changing brightness doesn't help (if it did, the camera would likely do that on its own).

The camera can't do that on its own when the user sets ISO above base rather than using Auto ISO.

What do you think? Is the above a reasonable overview? Have I left anything important out? Have I included anything that should be left out?

I suppose that the answers to those questions depend on the purpose of the overview. ;-)

Is there a better wording?

Almost always.

 

[Michael Fryd]

Michael Fryd wrote:
...

I talk about "visible image noise" because that;s the noise we see with our eyes. This avoids a discussion of absolute noise levels and Signal to Noise ratio.

...

You can avoid discussing Signal to Noise Ratio (SNR) but that is what determines whether noise is "visible".

Yes. But I wanted to leave that for a separate discussion. I don't want to bring in additional issues if i don't have to.

...

I'm sure it's a semantic fine-point.
Not bringing it into the discussion doesn't change the fact that you would be discussing "visible noise" without any foundation for making that objective rather than subjective.
If you want to start with a subjective conversation, that's you're business; I prefer to be more objective.

Perhaps it's a difference on our teaching style.

In my opinion the human brain can grasp some complicated concepts, even though it does not understand the science behind those concepts.

For instance, a child can easily learn that some noises are louder than others, without fully understanding sound perception, or that sounds are vibrations. In fact many people are not aware of the Equal Loudness Curve of human hearing (it turns out that a noise that it measurably louder than another may actually be softer)

Yet, without a technical understanding many have an idea of what is meant by one sound is louder than another. Certainly, many of us think we know when our neighbors are being too loud!

In terms of ISO, I think the minimum understanding necessary is to be able to see that some images look noisier than another. I don't think one needs to understand the nature of image noise in order to understand the relationship with ISO.

Remember, when it comes right down to it, ISO is not a major factor in image noise. The more important factor is exposure.

 

[dwalby]

I gave relatively brief answers the last time, and didn't word things too clearly, let me try again.

So here is how I view the topic, tell me where I am right or wrong:

and let's ignore the ISO-invariant aspect for the time being just to simplify the discussion.

1. The optimum strategy is to collect as much light as possible, since that will maximize SNR based on photon noise. Ideally, we'd expose as far to the right as possible without clipping highlights, at the lowest ISO possible (base ISO), collecting the maximum amount of light possible. This produces the best SNR overall.

2. if we are limited so that we can't achieve the goals of #1, for example, we can't slow the shutter speed low enough for whatever reason. Then as we increase ISO our signal is amplified to levels similar to #1, just below clipping, but at some ISO other than the base ISO.

Let's assume ISO 100 for case #1 and ISO 400 for case #2.

Case #2 has higher absolute output read noise than case #1 because the pre-amplification read noise was multiplied by 4x, and in case #1 that same noise was multiplied by 1x. The post-amplification read noise is the same in both cases. The signal levels of both cases are essentially the same, just below highlight clipping. So in my mind, case 2 has lower output SNR because of the amplified read noise component, and I don't think you disagree with that statement.

Now, I get that as you amplify the signal more with higher ISO settings, the post-amplification read noise stays constant. And at higher ISO/gain settings fewer electrons are required for a given ADC value, so the post-amp read noise (in ADC units) when expressed in terms of electrons becomes lower.

The math part doesn't confuse me, what I'm struggling with is how that has any practical application. Ideally, I'd like to change the exposure to get closer to case 1, since that will increase my SNR. But if I can't, then I'm going to select the ISO that provides the highest signal levels without clipping, it wouldn't make sense to do anything else. The only scenario I can think of is if a "properly" exposed image in case 2 would actually be at ISO 200 according to the camera's metering, but I have enough headroom to push it to ISO 400 and ETTR even further. In that case the SNR would improve between ISO 200 and 400, is that what we're talking about?

 

[Michael Fryd]

Thanks for you helpful comments.

...

On most modern cameras ISO serves three purposes:

1. It sets the target exposure for the metering system - with higher ISO settings, the metering system will suggest lower exposures (less light)

I'd suggest that this isn't (or in good photography usually should't be) a purpose, but rather an effect. Other than those rare cases where one actually wants to increase noise why would one purposefully set a target exposure lower than it could be?

Furthermore, since a photographer should usually want to get the largest exposure possible within the constraints of DOF control, motion blur control and clipping, it seems to me that using ISO to restrict exposure is a much less optimal way to effect expsoure than to use the more direct exposure controls: aperture and shutter.

Yes. "Purposes" may not be the best word.

Perhaps "changing the ISO affects three aspects of the digital process"?

1. I have mentioned that an ISO setting that is typical for the expected exposure is frequently the way to maximize quality for that exposure.

I'm not sure about that.

Here's what I was thinking. Not all cameras are ISO invariant. The inner workings of the camera may change (perhaps only a little) when you change the ISO setting. The change is designed to improve image quality at exposure that the meter would suggest for that ISO.

Thus if you have to use an exposure that would traditionally require ISO 800 film, you should set the camera to ISO 800. In that way, the camera knows the general magnitude of the exposure to expect, and can configure itself accordingly.

You could try to trick the camera by setting it for ISO 400 and boosting a stop, but that might provide a worse result if at ISO 400 the camera picks a different configuration that is optimized for a higher exposure.

 

[Iliah Borg]

The inner workings of the camera may change (perhaps only a little) when you change the ISO setting

1 stop change in clipping point per 1 ISO stop, you call it a little?

Thus if you have to use an exposure that would traditionally require ISO 800 film, you should set the camera to ISO 800.

No, for the reasons that must be obvious for anybody going to teach what ISO speed is.

In that way, the camera knows

Cameras know nothing.

 

[Michael Fryd]

I gave relatively brief answers the last time, and didn't word things too clearly, let me try again.

So here is how I view the topic, tell me where I am right or wrong:

and let's ignore the ISO-invariant aspect for the time being just to simplify the discussion.

Many cameras vary only a little with ISO. You can get reasonable results by modeling them as ISO-invariant.

1. The optimum strategy is to collect as much light as possible, since that will maximize SNR based on photon noise. Ideally, we'd expose as far to the right as possible without clipping highlights, at the lowest ISO possible (base ISO), collecting the maximum amount of light possible. This produces the best SNR overall.

The ISO setting has very little (if anything) to do with the collection of light at a given exposure. Ideally, your want to maximize exposure without maxing out any pixels you care about. With most modern DSLRs, the exposure that yields this result is the same for any ISO setting.

If the "ideal" exposure is f/4 at 1/60 at ISO 100, you will likely get almost the same data in the raw file if you shot f/4 at 1/60 at ISO 800.

Either the ISO 100 or ISO 800 image will yield similar Signal to Noise ratios as they both received the same signal and the same shot noise. The ISO 800 image won't blow out additional pixels, as the ISO doesn't affect the raw values from the sensor.

2. if we are limited so that we can't achieve the goals of #1, for example, we can't slow the shutter speed low enough for whatever reason. Then as we increase ISO our signal is amplified to levels similar to #1, just below clipping, but at some ISO other than the base ISO.

Let's assume ISO 100 for case #1 and ISO 400 for case #2.

Case #2 has higher absolute output read noise than case #1 because the pre-amplification read noise was multiplied by 4x, and in case #1 that same noise was multiplied by 1x. The post-amplification read noise is the same in both cases. The signal levels of both cases are essentially the same, just below highlight clipping. So in my mind, case 2 has lower output SNR because of the amplified read noise component, and I don't think you disagree with that statement.

Now, I get that as you amplify the signal more with higher ISO settings, the post-amplification read noise stays constant. And at higher ISO/gain settings fewer electrons are required for a given ADC value, so the post-amp read noise (in ADC units) when expressed in terms of electrons becomes lower.

But most cameras don't work that way.

The amplification is pretty similar at ISO 100 and ISO 800. The difference is how we interpret that data.

If a particular sensor pixel records a value of 12,543 at ISO 100, it will likely record a very similar value at ISO 800.

The math part doesn't confuse me, what I'm struggling with is how that has any practical application. Ideally, I'd like to change the exposure to get closer to case 1, since that will increase my SNR. But if I can't, then I'm going to select the ISO that provides the highest signal levels without clipping, it wouldn't make sense to do anything else. The only scenario I can think of is if a "properly" exposed image in case 2 would actually be at ISO 200 according to the camera's metering, but I have enough headroom to push it to ISO 400 and ETTR even further. In that case the SNR would improve between ISO 200 and 400, is that what we're talking about?

The practical application is that you maximize image quality by maximizing light captured.

If you can't maximize light capture, then capture as much as you can, and set the ISO to yield a normal histogram (or such that the camera's meter agrees with your exposure).

By setting the camera's ISO to match your exposure, you have told the camera about how much light to expect. If there is any internal camera optimizations for low light, these will kick in with the ISO setting. If there are variable amounts of available optimizations, you will get the amount that the camera manufacturer felt would work best for that exposure.

If you set the ISO to high, you may find that the camera kicks in optimizations that don't work well with your bright areas. If you set the ISO too low, the camera may omit optimizations that help in low light.

 

[Iliah Borg]

The amplification is pretty similar at ISO 100 and ISO 800.

^^^^

It's not funny.

 

[Michael Fryd]

The inner workings of the camera may change (perhaps only a little) when you change the ISO setting

1 stop change in clipping point per 1 ISO stop, you call it a little?

A bit earlier in this thread I posted images taken at ISO 200, 400, 800, and 1600. I metered at ISO 200, and used the same exposure for all four images. Changing the ISO did not change the clipping point in the raw data. Bright areas in the ISO 100 capture, were not clipped in the ISO 1600 capture.

Thus if you have to use an exposure that would traditionally require ISO 800 film, you should set the camera to ISO 800.

No, for the reasons that must be obvious for anybody going to teach what ISO speed is.

Thanks for providing a helpful correction rather than simply alleging that someone is incorrect.

In that way, the camera knows

Cameras know nothing.

I guess we disagree on whether anthropomorphism has a place in education.

For instance I might say that when you press a certain button on the remote, the TV "knows" you want it to change channels. While this may not be technically correct, it would be readily understood by many.

 

[Michael Fryd]

The amplification is pretty similar at ISO 100 and ISO 800.

^^^^

It's not funny.

--
http://www.libraw.org/

Assuming the same exposure and an ISO invariant camera, how big of a difference would you expect to see in the data captured at a particular sensor pixel between a capture with the camera set at ISO 100, and a capture with the camera set to ISO 800?

Obviously, there would be a significant difference in the camera produced preview and/or JPEG, but the context here was the data captured by the sensor.

 

[Iliah Borg]

The inner workings of the camera may change (perhaps only a little) when you change the ISO setting

1 stop change in clipping point per 1 ISO stop, you call it a little?

A bit earlier in this thread I posted images taken at ISO 200, 400, 800, and 1600. I metered at ISO 200, and used the same exposure for all four images. Changing the ISO did not change the clipping point in the raw data

You are attempting to teach things you do not understand yourself. More than few people here tried to explain you how ISO speed setting works, for two years already. You are not making any effort to understand.

--
http://www.libraw.org/

 

[Iliah Borg]

The amplification is pretty similar at ISO 100 and ISO 800.

^^^^

It's not funny.

Assuming the same exposure and an ISO invariant camera, how big of a difference would you expect to see in the data captured at a particular sensor pixel between a capture with the camera set at ISO 100, and a capture with the camera set to ISO 800?

Up to 3 stops more data at ISO 100.

--
http://www.libraw.org/

 

[FingerPainter]

I gave relatively brief answers the last time, and didn't word things too clearly, let me try again.

So here is how I view the topic, tell me where I am right or wrong:

and let's ignore the ISO-invariant aspect for the time being just to simplify the discussion.

1. The optimum strategy is to collect as much light as possible, since that will maximize SNR based on photon noise.

Yes. It will also have no direct effect on the noise added by the camera. However, as the exposure goes up, the amount of shot noise also goes up, making the extra noise added by the camera less and less significant as a part of the total noise. The signal is the amount of light captured. SInce the shot noise goes up with the square root of the signal, the signal to noise ratio declines and the image looks less noisy, even though it has more total noise.

Ideally, we'd expose as far to the right as possible without clipping highlights,

OK.

at the lowest ISO possible (base ISO),

We use the lowest possible ISO if we have already got the maximum exposure that will not blow highlights.

collecting the maximum amount of light possible.

OK

This produces the best SNR overall.

Not quite. It produces the best ratio of signal to shot noise. It may not produce the best ratio of signal to total noise. In cases where the highest exposure we can get is below the theoretical exposure that would have been the hightest not to blow highlights, we can often reduce the ratio of signal to camera-added noise, without affecting the ratio of signal to shot noise, by increasing ISO.

2. if we are limited so that we can't achieve the goals of #1, for example, we can't slow the shutter speed low enough for whatever reason. Then as we increase ISO our signal is amplified to levels similar to #1, just below clipping, but at some ISO other than the base ISO.

Yes, as long as the ISO increase is implemented through analog gain. The signal is amplified. The shot noise is amplified by the same amount. Any noise added by the camera before the gain stage is also amplified by the same amount. So the signal to noise ratio for all that noise remains the same. However, any noise added by the camera after the gain stage is not amplified. So the SNR for that particular noise is reduced: the signal was amplified, but that portion of noise was not. When you reduce the SNR for one component of noise, while leaving the SNR for all other noise the same, the SNR for total noise is also reduced.

Let's assume ISO 100 for case #1 and ISO 400 for case #2.

OK.

Just to clarify, in case 2, the signal level before any amplification due to an ISO increase is 1/4 what it was in case 1, right?

Case #2 has higher absolute output read noise than case #1 because the pre-amplification read noise was multiplied by 4x, and in case #1 that same noise was multiplied by 1x.

OK

The post-amplification read noise is the same in both cases.

OK

The signal levels of both cases are essentially the same,

Post amplification? Yes. Pre-amplification? No.

just below highlight clipping. So in my mind, case 2 has lower output SNR because of the amplified read noise component, and I don't think you disagree with that statement.

It has a lower SNR because some of the noise wasn't increased by a factor of four but all of the signal was.

Now, I get that as you amplify the signal more with higher ISO settings, the post-amplification read noise stays constant.

Yes

And at higher ISO/gain settings fewer electrons are required for a given ADC value,

I'm not sure about this switch to electrons. Is it equivalent to voltage? If so, fewer electrons coming off the sensor, yes. Not fewer going to the ADC.

so the post-amp read noise (in ADC units) when expressed in terms of electrons becomes lower.

How does it diminish? It is the same as it always was. It wasn't amplified but it wasn't reduced either. It is smaller relative to the amplified signal.

The math part doesn't confuse me, what I'm struggling with is how that has any practical application. Ideally, I'd like to change the exposure to get closer to case 1, since that will increase my SNR. But if I can't, then I'm going to select the ISO that provides the highest signal levels without clipping, it wouldn't make sense to do anything else. The only scenario I can think of is if a "properly" exposed image in case 2 would actually be at ISO 200 according to the camera's metering, but I have enough headroom to push it to ISO 400 and ETTR even further. In that case the SNR would improve between ISO 200 and 400, is that what we're talking about?

I don't think so.

Let's try an example with numbers. Let's say that highlights will be blown at 420. Noise added before the gain stage is 4 and after the gains stage is 1.

In case A, where we get maximum exposure at ISO 100, we get a signal of 400. Shot noise is therefore 20 (Sqrt(400)).

Noise components are added in quadrature. So going into the ADC we have signal of 400 and noise of sqrt(20^2 + 4^2 + 1^2) = sqrt (400 + 16 +1) = 20.42. This gives us an SNR of 19.588.

In the second case, exposure is only 1/4 what it was in case 1, so signal is 100, and shot noise is 10. If we do not raise ISO to 400, what goes to the ADC is a signal of 100 and total noise of sqrt(10^2 + 4^2 + 1^2) = sqrt(100 + 16 + 1) = 10.82. This gives an SNR of 9.245. As expected, this is a bit under 1/2 of case 1.

Now, if we increase ISO to 400 through analog gain, then the signal is amplified to 400, the shot noise is amplified to 40, pre-gain camera-added noise is amplified to 16 and the post-gain noise is still 1. Total noise is sqrt(40^2 + 16^2 + 1^2) = sqrt(1600 + 256 + 1) = 9.282.

The SNR has increased because of the ISO increase, but not by very much. For an increase in ISO to have a significant effect on SNR, the camera-added noise needs to be close to as large as the shot noise or larger (only happens in very low exposures), and the noise added after the gain stage has to be a large component of all camera added noise (as it was for instance in older Canon systems).

 

[gollywop]

I gave relatively brief answers the last time, and didn't word things too clearly, let me try again.

So here is how I view the topic, tell me where I am right or wrong:

and let's ignore the ISO-invariant aspect for the time being just to simplify the discussion.

1. The optimum strategy is to collect as much light as possible, since that will maximize SNR based on photon noise. Ideally, we'd expose as far to the right as possible without clipping highlights, at the lowest ISO possible (base ISO), collecting the maximum amount of light possible. This produces the best SNR overall.

2. if we are limited so that we can't achieve the goals of #1, for example, we can't slow the shutter speed low enough for whatever reason. Then as we increase ISO our signal is amplified to levels similar to #1, just below clipping, but at some ISO other than the base ISO.

Let's assume ISO 100 for case #1 and ISO 400 for case #2.

If I understand you correctly, you're not playing the proper game. Your case #1 (again, if I understand you correctly) is ETTR (assuming the camera has base ISO 100); i.e., at ISO 100 you were able to increase exposure all the way to just clipping. This is ETTR.

Your case #2, by contrast, is 2 stops short of ETTR, requiring ISO 400 to achieve "just clipping."

This is not the comparison I am making. Rather, the question is, in a case like #2 where your exposure is limited to less than ETTR (by 2 stops in this case), should you take the shot at ISO 100 and brighten it 2 stops during processing, or should you do the brightening in camera with the camera's ISO control?

If the camera is ISO-variant (not an issue that can be ignored), say up to ISO 800 or 1600, then you will do better to do the brightening in-camera, since for the given exposure, the accompanying read noise will be less at ISO 400 than at ISO 100. This will give you raw data exhibiting better overall s/n – but of particular significance in the highly shadowed regions.

Case #2 has higher absolute output read noise than case #1 because the pre-amplification read noise was multiplied by 4x, and in case #1 that same noise was multiplied by 1x. The post-amplification read noise is the same in both cases. The signal levels of both cases are essentially the same, just below highlight clipping. So in my mind, case 2 has lower output SNR because of the amplified read noise component, and I don't think you disagree with that statement.

Your case #1 has 4 times the light (signal) and, hence, twice the s/n of case #2 just because of the exposure alone. In this situation, the effect of ISO on read noise is a most uninteresting question.

Now, I get that as you amplify the signal more with higher ISO settings, the post-amplification read noise stays constant. And at higher ISO/gain settings fewer electrons are required for a given ADC value, so the post-amp read noise (in ADC units) when expressed in terms of electrons becomes lower.

The math part doesn't confuse me, what I'm struggling with is how that has any practical application. Ideally, I'd like to change the exposure to get closer to case 1, since that will increase my SNR. But if I can't, then I'm going to select the ISO that provides the highest signal levels without clipping, it wouldn't make sense to do anything else. The only scenario I can think of is if a "properly" exposed image in case 2 would actually be at ISO 200 according to the camera's metering, but I have enough headroom to push it to ISO 400 and ETTR even further. In that case the SNR would improve between ISO 200 and 400, is that what we're talking about?

--
gollywop
I am not a moderator or an official of dpr. My views do not represent, or necessarily reflect, those of dpr.
http://g4.img-dpreview.com/D8A95C7DB3724EC094214B212FB1F2AF.jpg

 

[FingerPainter]

Thanks for you helpful comments.

...

On most modern cameras ISO serves three purposes:

1. It sets the target exposure for the metering system - with higher ISO settings, the metering system will suggest lower exposures (less light)

I'd suggest that this isn't (or in good photography usually should't be) a purpose, but rather an effect. Other than those rare cases where one actually wants to increase noise why would one purposefully set a target exposure lower than it could be?

Furthermore, since a photographer should usually want to get the largest exposure possible within the constraints of DOF control, motion blur control and clipping, it seems to me that using ISO to restrict exposure is a much less optimal way to effect expsoure than to use the more direct exposure controls: aperture and shutter.

Yes. "Purposes" may not be the best word.

Perhaps "changing the ISO affects three aspects of the digital process"?

I think that is better.

However, don't lose sight of the fact that ISO is there for a purpose, and that purpose is to get a certain brightness level from a given exposure.

1. I have mentioned that an ISO setting that is typical for the expected exposure is frequently the way to maximize quality for that exposure.

I'm not sure about that.

Here's what I was thinking. Not all cameras are ISO invariant. The inner workings of the camera may change (perhaps only a little) when you change the ISO setting.

OK

The change is designed to improve image quality at exposure that the meter would suggest for that ISO.

Maybe. It might help if you explained what sort of change you have in mind.

Also keep in mind that on digital cameras, meters and the EC dial target a brightness level, not an exposure. The camera may choose to achieve the brightness level by adjusting exposure, ISO or both. In fact, on film cameras, they also target a brightness level, but the only way they have of manipulating brightness is by adjusting exposure.

Thus if you have to use an exposure that would traditionally require ISO 800 film, you should set the camera to ISO 800. In that way, the camera knows the general magnitude of the exposure to expect, and can configure itself accordingly.

I think this is a non-optimal way of looking at it on digital cameras. Drop the film-centric thought process.

The meter reads the amount of light and the "autoexposure" system determines how much exposure plus ISO brightening it will take to achieve middle grey brightness from that amount of light (or some tone other than middle grey as directed by the EC setting and input from any evaluative metering component). The camera's autoexposure system then takes into account any settings made by the user and adjusts one or more of the the other settings to get the desired brightness from the measured amount of light. In M mode without Auto-ISO, it doesn't make settings adjustment, it just indicates the result the current settings will give relative to the desired brightness. In other modes, if Auto-ISO is not set, then the ISO setting chosen by the user determines the exposure value the camera will use. if Auto-ISO is set, the camera is free to use the largest exposure it can that gives the desired brightness. Only if the largest EV available is insufficient to achieve the target brightness will it increase ISO.

You could try to trick the camera by setting it for ISO 400 and boosting a stop, but that might provide a worse result if at ISO 400 the camera picks a different configuration that is optimized for a higher exposure.

Example?

 

[FingerPainter]

The inner workings of the camera may change (perhaps only a little) when you change the ISO setting

1 stop change in clipping point per 1 ISO stop, you call it a little?

A bit earlier in this thread I posted images taken at ISO 200, 400, 800, and 1600. I metered at ISO 200, and used the same exposure for all four images. Changing the ISO did not change the clipping point in the raw data. Bright areas in the ISO 100 capture, were not clipped in the ISO 1600 capture.

Then you had at least three stops of highlight headroom in the ISO 200 shot,

Try the same shots under the same light but give them all three stops more exposure. I think you'll find that at least the ISO 1600 shot will have clipped highlights, but the ISO 200 shot will not have clipped highlights.

 

[Iliah Borg]

The inner workings of the camera may change (perhaps only a little) when you change the ISO setting

1 stop change in clipping point per 1 ISO stop, you call it a little?

A bit earlier in this thread I posted images taken at ISO 200, 400, 800, and 1600. I metered at ISO 200, and used the same exposure for all four images. Changing the ISO did not change the clipping point in the raw data. Bright areas in the ISO 100 capture, were not clipped in the ISO 1600 capture.

Then you had at least three stops of highlight headroom in the ISO 200 shot,

Or he is using a camera that has ISO as a tag

Or his converter "recovers" highlights.

See, he is saying "Changing the ISO did not change the clipping point in the raw data." - does that mean he actually looked at raw data, or simply that he does not understand the difference between the histogram / clipping displayed in a raw converter and raw data?

 

[Michael Fryd]

...

The change is designed to improve image quality at exposure that the meter would suggest for that ISO.

Maybe. It might help if you explained what sort of change you have in mind.

Also keep in mind that on digital cameras, meters and the EC dial target a brightness level, not an exposure. The camera may choose to achieve the brightness level by adjusting exposure, ISO or both. In fact, on film cameras, they also target a brightness level, but the only way they have of manipulating brightness is by adjusting exposure.

Thus if you have to use an exposure that would traditionally require ISO 800 film, you should set the camera to ISO 800. In that way, the camera knows the general magnitude of the exposure to expect, and can configure itself accordingly.

I think this is a non-optimal way of looking at it on digital cameras. Drop the film-centric thought process.

The meter reads the amount of light and the "autoexposure" system determines how much exposure plus ISO brightening it will take to achieve middle grey brightness from that amount of light (or some tone other than middle grey as directed by the EC setting and input from any evaluative metering component). The camera's autoexposure system then takes into account any settings made by the user and adjusts one or more of the the other settings to get the desired brightness from the measured amount of light. In M mode without Auto-ISO, it doesn't make settings adjustment, it just indicates the result the current settings will give relative to the desired brightness. In other modes, if Auto-ISO is not set, then the ISO setting chosen by the user determines the exposure value the camera will use. if Auto-ISO is set, the camera is free to use the largest exposure it can that gives the desired brightness. Only if the largest EV available is insufficient to achieve the target brightness will it increase ISO.

I'm trying to think of a simpler way of saying all of this.

Maybe we are back to the "Brightness Triangle"? Where Aperture and Shutter speed are the camera settings that affect exposure (and noise) and ISO affects brightness?

I like to view it this way: The metering system measures the light and determines an exposure for the base ISO. This is the camera's idea (and I know that camera's don't really think) of what the ideal exposure should be. (Keep in mind the camera is not very smart, and you may not agree with its selected exposure).

Now in order to achieve that exposure we need to select a combination of shutter speed and aperture. Due to other constraints (DoF, motion, Lens limitations, etc) there may not be an available combination that provides sufficient exposure. In this case a lower exposure might be chosen, and the ISO indicates how far off we are from the base ISO exposure.

Depending on the camera settings, the camera or the user may be in control of one or more of these three.

To confuse the situation some photographers use a workflow left over from the film world. The photographer has an idea of what he needs for an image. He may be in a situation when maximizing DoF or motion freezing may be more important than minimizing noise. If the photographer is familiar with his gear, he may have a good feeling for how noisy an image will look if he selects a particular ISO and allows the camera to meter it. For example, the photographer may decide that the exposure typically associated with ISO 1600 has acceptable noise for his needs. He may then set the camera's ISO to 1600 and use one of the program modes to maximize his available depth of field or shutter speed while staying under his noise tolerance. While setting the ISO is certainly not the same thing as specifying the noise level you can tolerate, in many circumstances it provides a useful approximation. You might even hear a photographer saying "with this camera I need to shoot at ISO 800 or below in order to avoid too much noise"

It is the above behavior that leads many new photographers to believe that the ISO setting is the "noise" setting.

You could try to trick the camera by setting it for ISO 400 and boosting a stop, but that might provide a worse result if at ISO 400 the camera picks a different configuration that is optimized for a higher exposure.

Example?

Let's suppose that we get our desired brightness at a particular exposure with an in-camera JPEG at ISO 400. We would not improve the image by keeping the exposure the same, lowering the ISO to 200 and then increasing the brightness from the raw file by one stop.

If noise was caused by high ISO, then shooting at ISO 200 would result in lower noise, and we would get a better image by boosting.

If noise is caused by exposure, then the ISO 400 and ISO 200 images will have the same shot noise. The ISO 400 image might be slightly better because the camera may have internally optimized itself for a typical ISO 400 exposure.