ISO Invariance: an experiment

Two recent threads (this and this) have generated a lot of discussion of "ISO Invariance" and related matters. These discussions have been almost entirely theoretical and largely centred on dynamic range and noise.

However, the experimental approach ("suck it and see") raises a couple of extra points that seem to be often overlooked.

For example, here are two images with the same shutter speed and aperture, but differing ISO settings:




Taken at ISO 6400 and at the exposure indicated on the camera's meter. Default processing in Lightroom.




Taken at ISO 200 and the same exposure as the previous shot. Processed in Lightroom with Exposure +5.0

The first point I would like to make is that to do this experiment, you need to choose not only the camera, but also the software to do the raw processing. I haven't tried processing with anything other than Lightroom, but I would guess that it makes a difference what software is used. For instance, Lightroom allows a maximum of 5.0 stops of exposure compensation.

The second point is that it is obvious from the shots above that not only do the noise levels and noise characteristics differ in the two images, but the colour balance differs also. Both shots were processed at daylight white balance set in the camera, and then carried through to LR.

Tom Axford said:

Two recent threads (this and this) have generated a lot of discussion of "ISO Invariance" and related matters. These discussions have been almost entirely theoretical and largely centred on dynamic range and noise.

However, the experimental approach ("suck it and see") raises a couple of extra points that seem to be often overlooked.

For example, here are two images with the same shutter speed and aperture, but differing ISO settings:


Taken at ISO 6400 and at the exposure indicated on the camera's meter. Default processing in Lightroom.


Taken at ISO 200 and the same exposure as the previous shot. Processed in Lightroom with Exposure +5.0

The first point I would like to make is that to do this experiment, you need to choose not only the camera, but also the software to do the raw processing. I haven't tried processing with anything other than Lightroom, but I would guess that it makes a difference what software is used. For instance, Lightroom allows a maximum of 5.0 stops of exposure compensation.

The second point is that it is obvious from the shots above that not only do the noise levels and noise characteristics differ in the two images, but the colour balance differs also. Both shots were processed at daylight white balance set in the camera, and then carried through to LR.

Click to expand...

I conducted virtually the same experiment over a year ago with my camera which Bill Claff suggests is iso invariant within the iso range I used. (On my camera iso 200 is NOT within that invariant range so my low end iso was 400). And I seem to recall that my difference was closer to 4-stops than 5.

I used a different raw converter than you (the one provided by the camera manufacturer which is able to read and duplicate the camera settings). I too detected a color balance difference. But in my experiment the noise results seemed virtually identical.

Tom Axford said:

Two recent threads (this and this) have generated a lot of discussion of "ISO Invariance" and related matters. These discussions have been almost entirely theoretical and largely centred on dynamic range and noise.

However, the experimental approach ("suck it and see") raises a couple of extra points that seem to be often overlooked.

For example, here are two images with the same shutter speed and aperture, but differing ISO settings:


Taken at ISO 6400 and at the exposure indicated on the camera's meter. Default processing in Lightroom.


Taken at ISO 200 and the same exposure as the previous shot. Processed in Lightroom with Exposure +5.0

The first point I would like to make is that to do this experiment, you need to choose not only the camera, but also the software to do the raw processing. I haven't tried processing with anything other than Lightroom, but I would guess that it makes a difference what software is used. For instance, Lightroom allows a maximum of 5.0 stops of exposure compensation.

The second point is that it is obvious from the shots above that not only do the noise levels and noise characteristics differ in the two images, but the colour balance differs also. Both shots were processed at daylight white balance set in the camera, and then carried through to LR.

Click to expand...

For noise levels, it's hard to call. Maybe as much as a stop, but maybe less. Color is clearly different.

But the ISO invariant portion of that camera seems to start at ISO 800. Maybe if you compared ISO800 and ISO6400, they would look more or less the same:



Bil Claff might chime in and explain this better, but from what I'm seeing, you stop getting shadow improvement after ISO800. So that is where I'd be hunting if I were trying to test for ISO invariance.

Yes. After making my previous post I checked Bill Claff's charts for the OP's camera and see that the iso's he used were not even close to all being in an invariant range. And even within the iso range where they are somewhat invariant they are not as flat-line even as for my camera (a Nikon D500).

stevo23 said:

But the ISO invariant portion of that camera seems to start at ISO 800. Maybe if you compared ISO800 and ISO6400, they would look more or less the same:



Bil Claff might chime in and explain this better, but from what I'm seeing, you stop getting shadow improvement after ISO800. So that is where I'd be hunting if I were trying to test for ISO invariance.

Click to expand...

That's a perfectly valid point, but I wasn't trying to find ISO invariance, I was trying to see what difference it would make if I set ISO 200 when the camera meter indicated ISO 6400 (and then push-processed to compensate).

The chart above suggests that I shouldn't drop the ISO below 800 in this situation, and that is indeed what I usually do with this camera (E-M10 II).

Doug Haag said:

Yes. After making my previous post I checked Bill Claff's charts for the OP's camera and see that the iso's he used were not even close to all being in an invariant range. And even within the iso range where they are somewhat invariant they are not as flat-line even as for my camera (a Nikon D500).

Click to expand...

That's true - I noticed that too. I'm not sure how much one would notice that sawtooth behavior in actual experience, but you're absolutely right. Many cameras are very flat.

Tom Axford said:

stevo23 said:

But the ISO invariant portion of that camera seems to start at ISO 800. Maybe if you compared ISO800 and ISO6400, they would look more or less the same:



Bil Claff might chime in and explain this better, but from what I'm seeing, you stop getting shadow improvement after ISO800. So that is where I'd be hunting if I were trying to test for ISO invariance.

Click to expand...

That's a perfectly valid point, but I wasn't trying to find ISO invariance, I was trying to see what difference it would make if I set ISO 200 when the camera meter indicated ISO 6400 (and then push-processed to compensate).

Click to expand...

Sorry, didn't mean to put words in your mouth. I have to say, it's not too bad of a result.

Tom Axford said:

The chart above suggests that I shouldn't drop the ISO below 800 in this situation, and that is indeed what I usually do with this camera (E-M10 II).

Click to expand...

On my Fuji XP2, ISO 800 is a better setting than ISO300-400 it would seem.

For an illustration of the degree to which the D600/610 sensor is ISO invariant, visit: https://www.dpreview.com/forums/post/57213569

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot. Instead, what I see is the ISO 6400 shot being approximately one stop less noisy than the ISO 200 shot. The result seems consistent with Bill Claff's measurements (no surprise there).

FingerPainter said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot.

Click to expand...

I would have interpreted the "exposure triangle" model to suggest that the ISO 200 shot would need 5 stops more exposure by adjusting either shutter speed, aperture or both!

I wasn't aware that the "exposure triangle" told us anything about noise.

FingerPainter said:

Instead, what I see is the ISO 6400 shot being approximately one stop less noisy than the ISO 200 shot. The result seems consistent with Bill Claff's measurements (no surprise there).

Click to expand...

FingerPainter said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot. Instead, what I see is the ISO 6400 shot being approximately one stop less noisy than the ISO 200 shot. The result seems consistent with Bill Claff's measurements (no surprise there).

Click to expand...

Not sure what you're implying here. That at ISO200, the camera isn't ISO invariant?

Tom Axford said:

The second point is that it is obvious from the shots above that not only do the noise levels and noise characteristics differ in the two images, but the colour balance differs also. Both shots were processed at daylight white balance set in the camera, and then carried through to LR.

Click to expand...

Yes, one must be careful here. I don't know about other raw processors, but the LR/ACR default camera profiles all tend to be twisted, i.e., they make supposedly "desirable" color shifts with increased exposure and/or raw-processor brightness (incorrectly called exposure). This is described at

http://dcptool.sourceforge.net/Hue Twists.html.

Your experiments almost certainly play straight into this mess.

The issue can be greatly mitigated by using untwisted camera profiles. Custom profiles made, for example, using an appropriate color checker along with Adobe's DNG Profile Editor or QPCalibration are not only tailored to your specific camera, but are also untwisted. Alternatively, one can make use of the dcptool software (associated with the above link) to untwist the Adobe profiles.

Beyond that, as has already been noted, the E-M10II doesn't begin to become ISO-invariant till ISO-800 or 1600. See Claff's chart .

--
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 said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot.

Click to expand...

Why. The exposure settings are the same, the iso 200 shot is not receiving more light.

gollywop said:

Tom Axford said:

The second point is that it is obvious from the shots above that not only do the noise levels and noise characteristics differ in the two images, but the colour balance differs also. Both shots were processed at daylight white balance set in the camera, and then carried through to LR.

Click to expand...

Yes, one must be careful here. I don't know about other raw processors, but the LR/ACR default camera profiles all tend to be twisted, i.e., they make supposedly "desirable" color shifts with increased exposure and/or raw-processor brightness (incorrectly called exposure). This is described at

http://dcptool.sourceforge.net/Hue Twists.html.

Your experiments almost certainly play straight into this mess.

Click to expand...

Many thanks for that information. I suspected that something along those lines was happening, but I hadn't read a detailed explanation of it before.

I can see the benefit in twisting the colour profiles after the brightness processing has been done, but I wonder why Adobe does it before processing brightness? Perhaps it is just easier for them?

Tom Axford said:

FingerPainter said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot.

Click to expand...

I would have interpreted the "exposure triangle" model to suggest that the ISO 200 shot would need 5 stops more exposure by adjusting either shutter speed, aperture or both!

Click to expand...

And yet you didn't need 5 stops more exposure did you? Instead you supplied 5 stops of lightening (not exposure) in PP.

Actually most expositions of the exposure triangle are very poor at describing how much exposure is actually needed.

Tom Axford said:

I wasn't aware that the "exposure triangle" told us anything about noise.

Click to expand...

Nearly every exposition of the exposure triangle I have seen says that noise increases with ISO, or that increasing ISO increases the noise.

Tom Axford said:

Tom Axford said:

Instead, what I see is the ISO 6400 shot being approximately one stop less noisy than the ISO 200 shot. The result seems consistent with Bill Claff's measurements (no surprise there).

Click to expand...

Click to expand...

stevo23 said:

FingerPainter said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot. Instead, what I see is the ISO 6400 shot being approximately one stop less noisy than the ISO 200 shot. The result seems consistent with Bill Claff's measurements (no surprise there).

Click to expand...

Not sure what you're implying here. That at ISO200, the camera isn't ISO invariant?

Click to expand...

I'm implying that the OP's experiment proves (once again) the typical exposition of the "exposure triangle" to be incorrect.

confused circle said:

FingerPainter said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot.

Click to expand...

Why. The exposure settings are the same, the iso 200 shot is not receiving more light.

Click to expand...

The typical explanation of the exposure triangle tells us that noise depends on ISO setting, and that as ISO increases so does noise.

In contrast, the "total light" explanation, on its own, would have us expecting the noise to be the same between both sample image, because as you say, they are made with the same amount of light.

Fortunately, the "total light" explanation is not usually presented in isolation of the potential SNR benefits of increased ISO. In fact the noisiness is worse in the lower ISO sample image, because the increase in ISO has increased the SNR.

FingerPainter said:

Tom Axford said:

FingerPainter said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot.

Click to expand...

I would have interpreted the "exposure triangle" model to suggest that the ISO 200 shot would need 5 stops more exposure by adjusting either shutter speed, aperture or both!

Click to expand...

And yet you didn't need 5 stops more exposure did you? Instead you supplied 5 stops of lightening (not exposure) in PP.

Actually most expositions of the exposure triangle are very poor at describing how much exposure is actually needed.

Click to expand...

?

FingerPainter said:

FingerPainter said:

I wasn't aware that the "exposure triangle" told us anything about noise.

Click to expand...

Nearly every exposition of the exposure triangle I have seen says that noise increases with ISO, or that increasing ISO increases the noise.

Click to expand...

I wasn't aware that exposure triangle tells us about noise. That's an semi-unrelated consequence I think.

confused circle said:

FingerPainter said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot.

Click to expand...

Why. The exposure settings are the same, the iso 200 shot is not receiving more light.

Click to expand...

However, this requires you to manually set some combination of aperture and time, and then change the ISO number without changing either of these. Which is perfectly possible, but it's not the way we normally work.

If you are shooting in A or S mode, for instance, then changing the ISO number will cause the camera to change one of the exposure settings. If you are in manual mode, there will be some kind of marker that tells you the shot will be "wrongly" exposed, and normally you would change speed or aperture to re-centre the marker.

Also, the view on the LCD or EVF will look too dark or too light if you change only the ISO number and keep the exposure unchanged.

So I think these tests are interesting and worth doing, but don't match normal camera usage. They may confuse inexperienced users even more than the "exposure triangle" does.

FingerPainter said:

stevo23 said:

FingerPainter said:

The "exposure triangle" model, as usually taught, would have us expect the ISO 6400 shot to be about 5 stops noisier than the ISO 200 shot. Instead, what I see is the ISO 6400 shot being approximately one stop less noisy than the ISO 200 shot. The result seems consistent with Bill Claff's measurements (no surprise there).

Click to expand...

Not sure what you're implying here. That at ISO200, the camera isn't ISO invariant?

Click to expand...

I'm implying that the OP's experiment proves (once again) the typical exposition of the "exposure triangle" to be incorrect.

Click to expand...

Maybe the typical exposition of the exposure triangle starts with an incorrect premise? I've not really followed exposure triangle discussions of late.

Noise is noise and does tend to go up with ISO, but exposure should only be about getting correct exposure - 18% grey etc. etc.

Noise is another discussion altogether. For instance: Shots at ISO800 on my Fuji X-Pro2 actually seem to display less noise than shots at ISO400-ISO600 due to the dual gain structure. That would tend to invalidate that typical exposition.