Is Changing ISO Just Adjusting for Incorrect Sensor Exposure?

toktik

Senior Member
Messages
1,409
Solutions
38
Reaction score
1,037
After about 20 years of not engaging in much photography, I returned to it about a year ago via Fujifilm. While refreshing my knowledge of photography, I have seen explanations that ISO is part of the "exposure triangle," but that doesn't seem correct to me.

Since exposure happens at the sensor, before ISO gain is applicable, it seems that shutter speed, aperture size, and available light are the components that impact exposure. Isn't changing ISO just a way to help compensate for over or under exposure that happens at the sensor?

From what I have learned about ISO, through reading and use, it seems to function similar to an audio amplifier while available light, shutter speed, and aperture work together to produce the music track (image) being made audible by the amplifier.

Am I misunderstanding something about the function of ISO gain?
 
It's interesting that higher ISO setting for lower exposure actually reduces noise (generally) than lower exposure brightened in PP, which seems to surprise many.
Same exposure, once with ISO speed setting up in camera, once with ISO speed (lightness) up in conversion, and the one with higher ISO speed set in the camera arrives with better SNR in shadows if raising ISO speed setting in the camera results in materially less read noise. Trivial ;)
I get it; what I don't get is why under ISO and brighten in PP? Even if it's "almost" as good as increasing ISO in camera?
I do it when the scene dynamic range exceeds the dynamic range the camera offers at "optimal" ISO, because setting a lower ISO and lightening in a converter brings less noise than setting a lower exposure in order to protect the highlights.
Thanks for that - still, we'd do what you describe above regardless whether camera is ISO invariant or not, no?

Does ISO invariancy affect either of the following?

Increase the exposure (within DOF and motion blur limits) until high light is near saturation.

If the exposure can/should not be increased, then increase the ISO until high light is near saturation.
 
Last edited:
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
There is no benefit and I'm not a fan of deliberately using ISO Invariance in that way.
Higher analog gain from raising the ISO setting always improves shadow performance but that improvement is rarely visible when it's over an ISO Invariant range.
So, in those cases not visibly harmful but not helpful (at all).
Hope you are not getting sick of me ;-)

How much does quantization noise affected by "analog gain"? Not visible? Visible at 100% crop?
 
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
Yes. For example, if I'm shooting with my D500 and anticipate a need to push image lightness by a full stop or more in post, may target exposure settings that work well with an ISO of 400 or higher. ISO 400 is where the D500's dual gain kicks in. I believe the same is true for the Z7 while the Z6's dual gain is at about ISO 800. The X-T3 (and presumably X-T30) also has a dual gain at about ISO 800.

This is a strategy that can work well for birds in flight or sports; genres for which fast shutter speeds can easily push ISOs to 400+ even with in good light with the lens wide open. For landscape photography - especially, if I'm on a tripod - I often opt to use settings to maximize exposure without blowing out highlights to take maximal advantage of the sensor's dynamic range. The sensor may not be invariant at base ISO but the noise advantage of maximal exposure ends of being of greater value to me. Also, I can often opt to bracket exposures and later blend them in post to bring out maximum detail in shadows, illuminated landscape and sky while doing minimal lightness adjustment.

Genre, subject and light all play a role in whether or not it's to one's benefit to maximize exposure or to take advantage of ISO invariance.
If you feel the need for a stop of highlight protection then regardless of the camera I'd simply set ISO Auto (in your case starting at ISO 400) and 1 stop of exposure compensation.
This strategy has nothing to do with ISO Invariance.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )
 
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
There is no benefit and I'm not a fan of deliberately using ISO Invariance in that way.
Higher analog gain from raising the ISO setting always improves shadow performance but that improvement is rarely visible when it's over an ISO Invariant range.
So, in those cases not visibly harmful but not helpful (at all).
Hope you are not getting sick of me ;-)
Not at all.
How much does quantization noise affected by "analog gain"? Not visible? Visible at 100% crop?
There is no meaningful affect; period.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )
 
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
Bird in flight and sports photography are two genres where 1) you're typically limited to working with one exposure, 2) using shutter speeds that - even in good light with the lens wide open - limit the volume of light captured during a shutter actuation and 3) using an ISO to protect highlights (white bird heads/bodies or white unis). In these scenarios, I often push a stop in post (rarely more than 1.5-2 stops) to optimize lightness and recover detail in the darker areas of the scene.

A high dynamic range landscape scene for which merging multiple exposures is not an option, is another scenario where shooting at or near base ISO but choosing exposure settings to protect highlights can result in pushing the shadows by 1-2 stops in post. This is a scenario where you may not be taking advantage of ISO invariance so much as the dynamic range of the sensor at max exposure.
If camera is not adding "analog gain", it seems it's in the "expanded ISO" mode; and if camera is adding "analog gain", the result will be still better but certainly never worse.

If the concern is clipping high light with high ISO, then the ISO is too high, no?
In most cases, yes, I would choose to reduce ISO as opposed to reducing exposure. However, there are rare occasions where I may anticipate needing to really push the shadows in a single exposure in post. In this circumstance, I may opt to reduce exposure - assuming the increased depth of field or faster shutter speed does not compromise the aesthetics of the shot - in order to use an ISO that's within the invariant range for a particular sensor.

Dual gain doesn't kick in in some cameras until ISO 800 or higher. For example, the D5 is a camera that really shines above about ISO 1600. If I were shooting fast action with a D5, I'd not hesitate to use a fast shutter speed, close down the aperture a tad for optimal sharpness and ISO 3200-6400. That sensor is just warming up in that exposure range.
Is it possible to plot "quantization noise", even if read noise is flat?
--
Bill Ferris Photography
Flagstaff, AZ
 
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
There is no benefit and I'm not a fan of deliberately using ISO Invariance in that way.
Higher analog gain from raising the ISO setting always improves shadow performance but that improvement is rarely visible when it's over an ISO Invariant range.
So, in those cases not visibly harmful but not helpful (at all).
Hope you are not getting sick of me ;-)
Not at all.
Thanks for that :-)
How much does quantization noise affected by "analog gain"? Not visible? Visible at 100% crop?
There is no meaningful affect; period.
Now you are going to get sick of me :-O

When I studied ADC, quantization noise was a big part; how could it not have any affect?
 
Last edited:
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
Bird in flight and sports photography are two genres where 1) you're typically limited to working with one exposure, 2) using shutter speeds that - even in good light with the lens wide open - limit the volume of light captured during a shutter actuation and 3) using an ISO to protect highlights (white bird heads/bodies or white unis). In these scenarios, I often push a stop in post (rarely more than 1.5-2 stops) to optimize lightness and recover detail in the darker areas of the scene.

A high dynamic range landscape scene for which merging multiple exposures is not an option, is another scenario where shooting at or near base ISO but choosing exposure settings to protect highlights can result in pushing the shadows by 1-2 stops in post. This is a scenario where you may not be taking advantage of ISO invariance so much as the dynamic range of the sensor at max exposure.
If camera is not adding "analog gain", it seems it's in the "expanded ISO" mode; and if camera is adding "analog gain", the result will be still better but certainly never worse.

If the concern is clipping high light with high ISO, then the ISO is too high, no?
In most cases, yes, I would choose to reduce ISO as opposed to reducing exposure. However, there are rare occasions where I may anticipate needing to really push the shadows in a single exposure in post. In this circumstance, I may opt to reduce exposure - assuming the increased depth of field or faster shutter speed does not compromise the aesthetics of the shot - in order to use an ISO that's within the invariant range for a particular sensor.

Dual gain doesn't kick in in some cameras until ISO 800 or higher.
Indeed, dual-gain adds a bit of wrinkle in selecting the "better" ISO. I recently got educated on this regarding PEN-F.
For example, the D5 is a camera that really shines above about ISO 1600. If I were shooting fast action with a D5, I'd not hesitate to use a fast shutter speed, close down the aperture a tad for optimal sharpness and ISO 3200-6400. That sensor is just warming up in that exposure range.
Is it possible to plot "quantization noise", even if read noise is flat?
 
...

But in the practical application I still think that "exposure triangle" is correct. You have 3 ways to expose a dark scene correctly: shutter speed at the cost of blur, aperture at the cost of DOF and ISO at the cost of noise. Or any combination of them.
Let's look at a practical example to see why the exposure triangle is wrong.

You are shooting BIF (mallards, not snow geese, so we don't have to worry about blown highlights) in the late evening with a D5 and an AF-S 200-500mm f/5,6G ED VR lens. Scene luminance is at about EV10 (for ISO 100). Your lens only opens up to f/5,6. You need at least 1/1000 shutter to avoid wing motion blur for the framing you are getting. You are not close enough to the subject to use flash effectively .

Your goal is to avoid wing motion blur while getting the least noise possible under the circumstances. DOF is not a consideration, and your lens is sufficiently sharp at f/5.6.

So what ISO should you set? What should your settings and processing be?

1) 1/1000 f/5.6 ISO 100, then lighten five stops in development
2) 1/1000 f/5.6 ISO 400, then lighten 3 stops in development.
3) 1/1000 f/5.6 ISO 3200, no need to lighten in development.

What does the exposure triangle tell us?

It tells us that if we increase ISO above base, our image will get noisier. So we should choose option 1.

In fact, we should choose option 3. If we do, we will have about 3.25 stops less noise in the shadows.
I've also heard what Banderras said about ISO != Gain. For me as an engineer that makes little to none sense, but there are apparently very knowledgeable people who support this theory.
It is not just a theory.

Gain is an analog boost to signal strength. There is nothing in the ISO standard that requires an ISO increase to be implemented by using gain. In fact, a few digital cameras, don;t use any variable gain to implement a change in ISO setting. Others only use variable gain to implement part of the range of ISO increases.
 
Last edited:
It's interesting that higher ISO setting for lower exposure actually reduces noise (generally) than lower exposure brightened in PP, which seems to surprise many.
Same exposure, once with ISO speed setting up in camera, once with ISO speed (lightness) up in conversion, and the one with higher ISO speed set in the camera arrives with better SNR in shadows if raising ISO speed setting in the camera results in materially less read noise. Trivial ;)
I get it; what I don't get is why under ISO and brighten in PP? Even if it's "almost" as good as increasing ISO in camera?
I do it when the scene dynamic range exceeds the dynamic range the camera offers at "optimal" ISO, because setting a lower ISO and lightening in a converter brings less noise than setting a lower exposure in order to protect the highlights.
we'd do what you describe above regardless whether camera is ISO invariant or not,
No, I don't do it regardless. I can refuse the shot, wait for light, recompose, use flash, use filters that lower the contrast, use a different camera, etc.
 
Technically you are perfectly right and you have explained it very good. You may have understood it even better than many experienced photographers :-)

But in the practical application I still think that "exposure triangle" is correct. You have 3 ways to expose a dark scene correctly: shutter speed at the cost of blur, aperture at the cost of DOF and ISO at the cost of noise. Or any combination of them.

I've also heard what Banderras said about ISO != Gain. For me as an engineer that makes little to none sense, but there are apparently very knowledgeable people who support this theory.
Let me try to explain myself a little better.

Analog amplification of the analog signal from the sensor is gain.

Remapping DNs after AD conversion is similar to the "levels" tool in photoshop, or histogram stretch in various other software. E.g. you have an 8 bit range, take a subset of that range, and map it to the full 8 bit range again.

Various outlets, e.g., fstoppers and Northrup, have compared high iso results in camera with pushed results in photoshop and found little difference, which supports the latter claim.

If you have a legitimate source about how iso is handled in reality in different camera designs, I would be very interested. So far what I read online is only theories and reverse engineering attempts. Moreover, I would not be surprised that canon and sony designs are different in this matter, as the DR graphs for canon and sony based cameras are very different in how they progress with ISOs.
ISO800 +4stop +50 midtone in DXO
ISO800 +4stop +50 midtone in DXO

ISO 6400 +50 midtone
ISO 6400 +50 midtone

Note the "SERVING SIZE" text...

It looks like ISO6400 applied more gain for better result. The difference is more noticible at full size viewing.
I can see the difference, and IMHO it's "little difference". Tony Northrup did this too and to him there was no difference. Fstoppers did this and found there is this difference.
The amount of difference will depend on the specific design of the camera. The more noise a camera adds after teh analog gain stage, the better the high ISO image will look.
That's not my point. I actually do not recommend shooting at base iso, I recommend proper in camera ISO setting based on the info online, because the output is better. My point is, people are too quick to attribute this difference to "analog" gain, when this is not a fact.

For example, what does RF signal has to do with this?
 
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
There is no benefit and I'm not a fan of deliberately using ISO Invariance in that way.
Higher analog gain from raising the ISO setting always improves shadow performance but that improvement is rarely visible when it's over an ISO Invariant range.
So, in those cases not visibly harmful but not helpful (at all).
Hope you are not getting sick of me ;-)
Not at all.
Thanks for that :-)
How much does quantization noise affected by "analog gain"? Not visible? Visible at 100% crop?
There is no meaningful affect; period.
Now you are going to get sick of me :-O

When I studied ADC, quantization noise was a big part; how could it not have any affect?
Because we're not down in the region of the ADC where quantization makes any visible difference. We have enough photon noise, even in the deep shadows, that it doesn't matter.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )
 
...

But in the practical application I still think that "exposure triangle" is correct. You have 3 ways to expose a dark scene correctly: shutter speed at the cost of blur, aperture at the cost of DOF and ISO at the cost of noise. Or any combination of them.
Let's look at a practical example to see why the exposure triangle is wrong.

You are shooting BIF (mallards, not snow geese, so we don't have to worry about blown highlights) in the late evening with a D5 and an AF-S 200-500mm f/5,6G ED VR lens. Scene luminance is at about EV10 (for ISO 100). Your lens only opens up to f/5,6. You need at least 1/1000 shutter to avoid wing motion blur for the framing you are getting. You are not close enough to the subject to use flash effectively .

Your goal is to avoid wing motion blur while getting the least noise possible under the circumstances. DOF is not a consideration, and your lens is sufficiently sharp at f/5.6.

So what ISO should you set? What should your settings and processing be?

1) 1/1000 f/5.6 ISO 100, then lighten five stops in development
2) 1/1000 f/5.6 ISO 400, then lighten 3 stops in development.
3) 1/1000 f/5.6 ISO 3200, no need to lighten in development.

What does the exposure triangle tell us?

It tells us that if we increase ISO above base, our image will get noisier. So we should choose option 1.

In fact, we should choose option 3. If we do, we will have about 3.25 stops less noise in the shadows.
I fully appreciate what you are saying above ;-) That ISO controls the noisiness is really the wrong part of "exposure triangle" along with increase in ISO increase exposure, should mean brightness (or lightness level)

However, in my practical application of "exposure triangle" is thus: image brightness from a camera is affected by F#, SS and ISO and there exists equivalent settings of F#, SS and ISO that result in similarly bright image. And choose the one combination that meets your DOF, motion blur and to some extent noisiness of the image you want.

Triangle means there are three parameters as in love triangle where there are three actors ;-)
I've also heard what Banderras said about ISO != Gain. For me as an engineer that makes little to none sense, but there are apparently very knowledgeable people who support this theory.
It is not just a theory.

Gain is an analog boost to signal strength. There is nothing in the ISO standard that requires an ISO increase to be implemented by using gain. In fact, a few digital cameras, don;t use any variable gain to implement a change in ISO setting. Others only use variable gain to implement part of the range of ISO increases.
 
I'm fascinated by all these amazing detailed posts about the inner workings of digital sensors but I always come away from reading these threads with no clear idea of how to use this information to make my images better.

I'd love for the people that seem to have a deep technical understanding of how digital sensors work to talk about practical strategies to apply the knowledge to actual shooting because I, apparently, am too dumb to figure it out on my own.
 
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
There is no benefit and I'm not a fan of deliberately using ISO Invariance in that way.
Higher analog gain from raising the ISO setting always improves shadow performance but that improvement is rarely visible when it's over an ISO Invariant range.
So, in those cases not visibly harmful but not helpful (at all).
Hope you are not getting sick of me ;-)
Not at all.
Thanks for that :-)
How much does quantization noise affected by "analog gain"? Not visible? Visible at 100% crop?
There is no meaningful affect; period.
Now you are going to get sick of me :-O

When I studied ADC, quantization noise was a big part; how could it not have any affect?
Because we're not down in the region of the ADC where quantization makes any visible difference. We have enough photon noise, even in the deep shadows, that it doesn't matter.
Sorry; I may have mixed up my question (bobn2 pointed this out many times) as the use of term "quantization noise". So if I may...

If the scene has10 stop DR and camera has 12 bit ADC, and if the signal is not amplified, the resulting image would have 2^10 tonal range. However, if the signal is amplified by 2 stops, the resulting image would have 2^12 tonal range.

Now if the 2^10 file is digitally multiplied by 4x, the tonal range can be extended to 2^12, however, the tonal distribution would be comb like, made up of only 2^10 different values.

I have attributed such difference "quantization noise", since the it's the difference of 10bit ADC vs 12bit ADC quantization noise.

Is there more correct term for having all the values from 0 to 2^12 vs having only 2^10 different values, even though spread out between 0 to 2^12?
 
Last edited:
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
There is no benefit and I'm not a fan of deliberately using ISO Invariance in that way.
Higher analog gain from raising the ISO setting always improves shadow performance but that improvement is rarely visible when it's over an ISO Invariant range.
So, in those cases not visibly harmful but not helpful (at all).
Hope you are not getting sick of me ;-)
Not at all.
Thanks for that :-)
How much does quantization noise affected by "analog gain"? Not visible? Visible at 100% crop?
There is no meaningful affect; period.
Now you are going to get sick of me :-O

When I studied ADC, quantization noise was a big part; how could it not have any affect?
Because we're not down in the region of the ADC where quantization makes any visible difference. We have enough photon noise, even in the deep shadows, that it doesn't matter.
Sorry; I may have mixed up my question (bobn2 pointed this out many times) as the use of term "quantization noise". So if I may...

If the scene has10 stop DR and camera has 12 bit ADC, and if the signal is not amplified, the resulting image would have 2^10 tonal range. However, if the signal is amplified by 2 stops, the resulting image would have 2^12 tonal range.

Now if the 2^10 file is digitally multiplied by 4x, the tonal range can be extended to 2^12, however, the tonal distribution would be comb like, made up of only 2^10 different values.

I have attributed such difference "quantization noise", since the it's the difference of 10bit ADC vs 12bit ADC quantization noise.

Is there more correct term for having all the values from 0 to 2^12 vs having only 2^10 different values, even though spread out between 0 to 2^12?
I'm not "sick of you" but I think we're wandering off topic.

Your notion of tonal range is incorrect. You need to think about distinguishable tones which involves paying attention to the Signal to Noise Ratio (SNR).
The DxOMark definition of Tonal Range is a reasonable place to start.
It's true that the more amplified signal has slightly better SNR but no so much so as to make a real visible difference in tonal range.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )
 
I'm fascinated by all these amazing detailed posts about the inner workings of digital sensors but I always come away from reading these threads with no clear idea of how to use this information to make my images better.

I'd love for the people that seem to have a deep technical understanding of how digital sensors work to talk about practical strategies to apply the knowledge to actual shooting because I, apparently, am too dumb to figure it out on my own.
You may be employing some strategies, already, without knowing it. A photographer who has been shooting with a camera for some time tends to develop a sense of which settings deliver a more pleasing image when doing certain kinds of photography. So, give yourself some credit. That said, it is to our benefit to be more intentional about these things.

I generally recommend starting by choosing a focal length and exposure settings that will deliver a shot with the framing (focal length), depth of field (f-stop) and degree of motion-freezing (shutter speed) you desire. These are the creative decisions that - along with choosing your perspective on the subject and when to press the shutter - go far in determining the quality of the resulting image.

With that out of the way, what ISO does the camera suggest you use? Of you're in full manual, the camera's recommended ISO will be that which produces a metering display parked at 0. The photo will be, on-meter. Now, in this situation, I'm treating ISO as the smoke indicating a possible fire. It's more an indicator than an end unto itself.

The question to keep in mind is, is ISO within a range where you typically get good results? If so, you're all set. If the ISO is outside this range, it's time to start considering your options. The acceptable range will vary, depending on the camera and your personal aesthetic. But generally speaking, if your settings will deliver enough light to the sensor to merit an ISO at base or 2-3 stops above base, you're in great shape. At 4-5 stops above base, things are still good. Higher than this and you probably want to invest some time rethinking your exposure settings.

As mentioned, the ISO range indicating exposures that deliver acceptable results will vary according to your tastes. For the sake of discussion, let's assume the ISO that corresponds to an on-meter display is higher than you like. Keeping in mind that ISO is an indicator (like smoke is to fire), the next step is to consider the f-stop and shutter speed settings. Will opening the aperture or slowing the shutter speed (either of which will deliver more light to the sensor) compromise your creative vision for the shot. If the answer to that question is, yes, there are other options to consider.

For one it's always an option to wait to press the shutter release until the shot improves. In this context, that may mean clouds passing to put more light on the subject, the subject moving from shade to sunlight, the sun moving to a position in the sky that more directly illuminates the subject, you moving to a new position from which the subject will be better illuminated.

Can you improve lighting by using the on-camera flash. an external speedlight, a reflector or other light source? If the answer to all these questions is, no, the best decision may be to wait...wait for another opportunity that comes with improved lighting.

Or, consider the option of making a different photo. For example, when I'm photographing elk grazing in a meadow at sunset, I'm keenly aware of slowing my shutter speed only so much. Once I've hit a threshold beyond which there's just no reasonable chance of freezing motion and capturing enough light for a good exposure, I'll sometimes go the other direction and start using shutter speeds so slow that I have to manually pan to follow the animal's motion during the length of the shutter actuation. If I'm successful, the resulting photo will feature a focused subject that's sharp in the face and an obviously blurry background. It's an aesthetically pleasing alternative to my standard approach of using a faster shutter speed to freeze motion.

By using a slow shutter speed, I'm emphasizing motion...calling attention to it. I'm also putting more light on the sensor, increasing exposure and improving the image IQ.

We could explore a host of possible scenarios and the variety of available solutions. But I'll stop, here, in the hope I've illustrated a thought process one can employ to get the shot and enough light to make the shot worth getting. In summary...
  • Start by choosing exposure settings that will deliver the photo you envision.
  • Check the ISO recommended by the camera as delivering an on-meter exposure.
  • In this context, ISO is an indicator we use to identify potential problems with the exposure settings. It's like smoke (indicator) to fire (problem).
  • If ISO indicates a possible problem, consider your options for changing exposure settings and go with what you consider the best adjustment or option.
  • Remember, it's OK to decline to press the shutter. It's also OK to improvise an alternative creative vision for the shot; one that works better with the available light.
 
If you repeated such an experiment with something that is far less ISO Invariant, like the D5, you'd get different results:

091ccc8c4b284b2b9deda144de3157be.jpg.png

I'm simply saying that one issue with ISO Invariance is that people sometimes blindly assume it will always work without checking their particular camera ; which is easily done as shown above.
What would be the benefit of "under ISO" even if PP is almost as good as "nominal in camera ISO", even if the camera is within "ISO invariant" range?
There is no benefit and I'm not a fan of deliberately using ISO Invariance in that way.
Higher analog gain from raising the ISO setting always improves shadow performance but that improvement is rarely visible when it's over an ISO Invariant range.
So, in those cases not visibly harmful but not helpful (at all).
Hope you are not getting sick of me ;-)
Not at all.
Thanks for that :-)
How much does quantization noise affected by "analog gain"? Not visible? Visible at 100% crop?
There is no meaningful affect; period.
Now you are going to get sick of me :-O

When I studied ADC, quantization noise was a big part; how could it not have any affect?
Because we're not down in the region of the ADC where quantization makes any visible difference. We have enough photon noise, even in the deep shadows, that it doesn't matter.
Sorry; I may have mixed up my question (bobn2 pointed this out many times) as the use of term "quantization noise". So if I may...

If the scene has10 stop DR and camera has 12 bit ADC, and if the signal is not amplified, the resulting image would have 2^10 tonal range. However, if the signal is amplified by 2 stops, the resulting image would have 2^12 tonal range.

Now if the 2^10 file is digitally multiplied by 4x, the tonal range can be extended to 2^12, however, the tonal distribution would be comb like, made up of only 2^10 different values.

I have attributed such difference "quantization noise", since the it's the difference of 10bit ADC vs 12bit ADC quantization noise.

Is there more correct term for having all the values from 0 to 2^12 vs having only 2^10 different values, even though spread out between 0 to 2^12?
I'm not "sick of you" but I think we're wandering off topic.

Your notion of tonal range is incorrect. You need to think about distinguishable tones which involves paying attention to the Signal to Noise Ratio (SNR).
The DxOMark definition of Tonal Range is a reasonable place to start.
Thanks. I'll look up.
It's true that the more amplified signal has slightly better SNR but no so much so as to make a real visible difference in tonal range.
 
RF gain is very much analogous to the "analog gain"; in that "analog gain", which is set by the ISO dial, increase the photographic sensitivity of a camera as the RF gain increases the "sensitivity" of a SW radio.

Brightening in PP does not increase the photographic sensitivity of a camera same way as increasing the audio volume does not increase the "sensitivity" of a SW radio.
RF gain and analogue gain are the same - they are both accomplished by amplifiers. When a RF receiver is supplied an antenna output, the signal is first filtered (by an analogue filter that isolates the frequency band of interest) and that is now made available to the ADC. If the output of the filter is not sufficient to allow the ADC to operate near full range then amplification is applied. It can be done by an AGC (automatic gain control - look it up ) or by an operator running a knob. However an amplifier is an active component and every active component has a noise figure - that is how much noise it adds to the resulting output signal and it is greater than zero. That's right every amplifier even those on CMOS sensor chips add noise. Once you turn up application (in the case of digital cameras with analogue gain) you increase the signal power and the noise power the same and add noise from the amplifier determined by the noise figure (look that up to). You amplify the signal plus noise in the input and add noise from the amplifier. The more amplification the more noise you normally add.

Now some claim the Sony sensors used by Fuji are ISO invariant. But lets take a look. If you simply grab what is there - fine. If you multiply the output of the detectors by 10 (about 3 stops) you multiply everything. When the signal is not in the proper range for a 14 bit ADC, say it is stops off and is only in the range of a 11 bit ADC, the signal as quantization noise of 3 bits. That is the signal noise ratio goes down because of quantization noise. So is it better to apply amplification by 3 stops and not have the quantization noise but have the amplifier noise. Or is it better to take the quantification noise hit and not have the amplifier noise?

With today's low noise amplifiers - no one designing an RF receiver would ever take a 6 dB hit per every bit not tickled most of the bits available.


This argument gets down to the absurd. Crank up the amplification to use the full range of the ADC while being careful not to saturate.

There is no free lunch.
 
I'm fascinated by all these amazing detailed posts about the inner workings of digital sensors but I always come away from reading these threads with no clear idea of how to use this information to make my images better.

I'd love for the people that seem to have a deep technical understanding of how digital sensors work to talk about practical strategies to apply the knowledge to actual shooting because I, apparently, am too dumb to figure it out on my own.
In this thread, as I am the most guilty of what you are saying (in number of posts and not the most technical knowledge), my sincere apologies for taking the discussion to the technical details rather than what to do with the camera settings...

I ask all those questions to the technically knowledgeable folks for the same reason you wonder "so, what should I do"? What should I do if my camera is ISO invariant, for example.

I came to the conclusion that digital sensor camera is not that different than shooting the slide film from yester years...

After all the discussions, I am now fairly confident that camera designers have implemented digital camera to mimic film camera and film ISO (ASA, back in the days) pretty well.

As it was with film camera, the brightness of the scene can vary more or less than the sensor/camera can record and more than the display media can display. Within that limitation, photographer must manage exposure and ISO setting to achieve desired result.

There are additional wrinkles but if the image looks too dark at the desired F# and SS, increase the ISO setting as you would load a faster film. The difference is with digital camera, you can do this frame by frame.

One of the more interesting Auto mode in digital camera is actually the Manual mode with Auto ISO. In this mode, EC will bias the Auto ISO rather than change the exposure.

At the end of the day, the biggest advantage of digital photography is that you can experiment, experiment and experiment without even waiting for the photos to come back. And without the film/print cost :-O How cool is that ;-)

Good luck!!
 
Last edited:
RF gain is very much analogous to the "analog gain"; in that "analog gain", which is set by the ISO dial, increase the photographic sensitivity of a camera as the RF gain increases the "sensitivity" of a SW radio.

Brightening in PP does not increase the photographic sensitivity of a camera same way as increasing the audio volume does not increase the "sensitivity" of a SW radio.
RF gain and analogue gain are the same - they are both accomplished by amplifiers. When a RF receiver is supplied an antenna output, the signal is first filtered (by an analogue filter that isolates the frequency band of interest) and that is now made available to the ADC. If the output of the filter is not sufficient to allow the ADC to operate near full range then amplification is applied. It can be done by an AGC (automatic gain control - look it up ) or by an operator running a knob. However an amplifier is an active component and every active component has a noise figure - that is how much noise it adds to the resulting output signal and it is greater than zero. That's right every amplifier even those on CMOS sensor chips add noise. Once you turn up application (in the case of digital cameras with analogue gain) you increase the signal power and the noise power the same and add noise from the amplifier determined by the noise figure (look that up to). You amplify the signal plus noise in the input and add noise from the amplifier. The more amplification the more noise you normally add. - bold added
A minor quibble: the signal (light in this case) includes the noise (shot noise). The variable gain amplifier adds its NF but it's same whether variable gain is set at 1x or 100x (within its specification).

That is why signal with higher gain (even though the noise of the signal is also amplified), is less noisy; SNR with 100x gain is higher than 1x gain, more so because the signal noise and amplifier NF add in quadrature.
Now some claim the Sony sensors used by Fuji are ISO invariant. But lets take a look. If you simply grab what is there - fine. If you multiply the output of the detectors by 10 (about 3 stops) you multiply everything. When the signal is not in the proper range for a 14 bit ADC, say it is stops off and is only in the range of a 11 bit ADC, the signal as quantization noise of 3 bits. That is the signal noise ratio goes down because of quantization noise. So is it better to apply amplification by 3 stops and not have the quantization noise but have the amplifier noise. Or is it better to take the quantification noise hit and not have the amplifier noise?
Yes, I looked up SQNR - Signal to Quantization Noise Ratio. For Uniformly distributed signal power, each additional bit improves SQNR by ~ 6dB.
With today's low noise amplifiers - no one designing an RF receiver would ever take a 6 dB hit per every bit not tickled most of the bits available.

https://classes.engineering.wustl.edu/ese488/Lectures/Lecture5a_QNoise.pdf

This argument gets down to the absurd. Crank up the amplification to use the full range of the ADC while being careful not to saturate.
For the same reason, I've wondering why "under ISO", whether the particular camera is ISO invariant or not. No camera manufacture ever designated their camera to be ISO invariant, either.
There is no free lunch.
 
Last edited:

Keyboard shortcuts

Back
Top