High ISO noise and ADC

[tony field]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

--
Charles Darwin: "ignorance more frequently begets confidence than does knowledge."
tony

 

[ThrillaMozilla]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I don't know for sure why he got the results he did, but his test was done very crudely, and probably wrong. I don't see how he could have made valid measurements the way he did it. We didn't see his raw data or his images, so it's hard to know exactly what he got or what he should have measured. For his camera there should not be much difference above ISO 1600.

I don't know for sure why he measured low values at low ISO, but he measured DN values on jpeg files, and I suspect that he made a mistake. He varied the ISO setting by a factor of 256. That's 8 stops, and that may cause serious numerical errors. I think it's possible that he clipped the low-ISO values, and in so doing, destroyed the record of noise. Alternatively, his method of exposure correction could have altered the numbers in a non-proportional way. The patches he measured may have looked noise free, but the method seems error-prone, and I'm not sure he actually measured noise with sufficient accuracy.

I don't want to discuss those further because the data is completely inadequate to say anything definitive. To put it bluntly, I don't believe his results.

EDIT: I still don't know what's wrong, but it's not credible that he gets lower noise by exposing for ISO 25600 but setting the ISO at 100. For his camera, saturation at ISO 25600 is 293 electrons and read noise is 0.611 electrons. At ISO 100, read noise is 6.77 electrons. At ISO 25600, S/N - 480 at most; at ISO 100 S/N is 43 at most, just considering read noise. Shot noise and fixed pattern noise are constant, so there is no way for total noise to be lower for ISO under the stated test conditions. There is way too much room for error in his methods, in my opinion.

EDIT 2: Under more realistic conditions, let's expose 1 step down to give some headroom, then expose a dark part of the image, another 5 steps down, instead of pure white. Then, considering read noise only, the S/N ratios would be 7.5 and 0.67 for ISO 25600 and 100, respectively. That picture would look horrible.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR.

That's to be expected. For your cameras the difference in read noise between ISO 3200 and 51200 is quite small. The difference between ISO 3200 and 100, on the other hand, is quite large.

It's worth making two other comments. First, read noise is only one source of noise, and it's usually visible only in the deep shadows. At higher exposures, shot noise (photon statistical noise) usually dominates. Second, be careful about expressing noise in units of digital numbers (DN). If you try to compare noise measured at ISO 100 with noised measured at ISO 200, be aware that the numbers you read in the files are usually not comparable at any two ISO values, because of some combination (1) a different voltage for each ISO value; (2) multiplication by a different scale factor; or (3) a different capacitor in the circuit. To get around this problem, signal and noise in the sensors are measured in actual physical units, namely electrons.

If you do the comparison correctly, many camera sensors do show significantly more read noise at low ISO values. Whether you will see the difference is another matter altogether. It just depends on the signal and noise levels. With modern cameras, read noise is tending to become insignificant in many cases.

A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

Maybe. Whether you can see the read noise depends on the camera, the exposure, and the brightness of the print.

 

[bobn2]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

For the Canon 1D and 5D the reduction was both large and also insignificant. Large because the read noise reduced by a factor of around 8 as the ISO was increased from 100 to 3200. Insignificant because the fixed pattern noise was so bad (and objectionable) that it rendered the whole question of the read noise level moot.

In pretty much all modern cameras the fixed pattern noise has been reduced to a level which makes it effectively not a problem. These cameras have a nice even random noise pattern which isn't visually objectionable.

The noise reduction is certainly there for most cameras, but isn't really a big issue. It's talked about mostly just as a counterpoint when someone says wrongly that increasing ISO increases noise.

Here are four Canon 5 cameras:





These are all at 100 ISO and visually not much different, even though the read noise is quite different.

According to Bill Claff's figures at 100 ISO the read noise (per pixel) is 27 e- for the 5D, 13 e- for the Mk IV, 11,5 for the Ds and 4.1 for the R5 (misleading because canon has applied involuntary NR to get there). Now let's look at 3200 ISO





We can see that the 5D is worse, and this tallies with Bill's figures, which are 4 e- for the 5D, 2 e- for the MK IV and Ds and 1.5 for the R5 (no NR this time). So, why can't we see all the extra noise at 100 ISO? The simple reason is that it's all below the black level. The 5D can collect about 50,000 electrons per pixel. Given a normal headroom that would mean that the white level for a shot exposed for 100 ISO would be around 28,000. That puts the read noise level about 1000 times lower, for an effective DR of 10 stops. That's quite enough to fit a full tonal scale from white to black. If, however, you'd set 100 ISO but exposed for 3200, that sets the white level at 875 electrons. Now the read noise is only 5 stops below the white, so is going to be above a normally set black level.

Also, from Bill's curves you'll find that the improvement above 3200 is pretty marginal in most cameras.

--
Is it always wrong
for one to have the hots for
Comrade Kim Yo Jong?

 

[Chris R-UK]

This may help from Lonely Speck, an astrophotography site:

https://www.lonelyspeck.com/how-to-find-the-best-iso-for-astrophotography-dynamic-range-and-noise/

Look for the comparison of the effect of aperture, shutter speed and ISO on noise about 1/3 of the way through.
--
Chris R

 

[ThrillaMozilla]

We can see that the 5D is worse, and this tallies with Bill's figures, which are 4 e- for the 5D, 2 e- for the MK IV and Ds and 1.5 for the R5 (no NR this time). So, why can't we see all the extra noise at 100 ISO? The simple reason is that it's all below the black level. The 5D can collect about 50,000 electrons per pixel. Given a normal headroom that would mean that the white level for a shot exposed for 100 ISO would be around 28,000. That puts the read noise level about 1000 times lower, for an effective DR of 10 stops. That's quite enough to fit a full tonal scale from white to black. If, however, you'd set 100 ISO but exposed for 3200, that sets the white level at 875 electrons. Now the read noise is only 5 stops below the white, so is going to be above a normally set black level.

Also, from Bill's curves you'll find that the improvement above 3200 is pretty marginal in most cameras.

Yes, exactly. There are good visual examples of this in the DPR camera reviews. They expose for a high ISO value but set the ISO at various values, then adjust the image brightness and compare the images. This is found in the section entitled "ISO invariance".

One scenario for this is an attempt to preserve headroom (high dynamic range) at low exposures, by exposing for a high ISO value and setting a low ISO value. Another very common scenario is to try to image deep shadows and strong highlights in the same picture. Some cameras are much better than others in that respect. That's why it's sometimes a big deal.

 

[tony field]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

For the Canon 1D and 5D the reduction was both large and also insignificant. Large because the read noise reduced by a factor of around 8 as the ISO was increased from 100 to 3200. Insignificant because the fixed pattern noise was so bad (and objectionable) that it rendered the whole question of the read noise level moot.

In pretty much all modern cameras the fixed pattern noise has been reduced to a level which makes it effectively not a problem. These cameras have a nice even random noise pattern which isn't visually objectionable.

The noise reduction is certainly there for most cameras, but isn't really a big issue. It's talked about mostly just as a counterpoint when someone says wrongly that increasing ISO increases noise.

Here are four Canon 5 cameras:



These are all at 100 ISO and visually not much different, even though the read noise is quite different.

According to Bill Claff's figures at 100 ISO the read noise (per pixel) is 27 e- for the 5D, 13 e- for the Mk IV, 11,5 for the Ds and 4.1 for the R5 (misleading because canon has applied involuntary NR to get there). Now let's look at 3200 ISO



We can see that the 5D is worse, and this tallies with Bill's figures, which are 4 e- for the 5D, 2 e- for the MK IV and Ds and 1.5 for the R5 (no NR this time). So, why can't we see all the extra noise at 100 ISO? The simple reason is that it's all below the black level. The 5D can collect about 50,000 electrons per pixel. Given a normal headroom that would mean that the white level for a shot exposed for 100 ISO would be around 28,000. That puts the read noise level about 1000 times lower, for an effective DR of 10 stops. That's quite enough to fit a full tonal scale from white to black. If, however, you'd set 100 ISO but exposed for 3200, that sets the white level at 875 electrons. Now the read noise is only 5 stops below the white, so is going to be above a normally set black level.

Also, from Bill's curves you'll find that the improvement above 3200 is pretty marginal in most cameras.

This clarifies some of my concerns.

In terms of "noise measurement", is the Photoshop standard deviation of a consistent size image patch an appropriate measure of noise as apparently done in D Cox original post? It is, in my estimation, at least a good comparative measure.

Although fixed pattern noise on the Canon 5D ll and 5D was easy to demonstrate, it did not actually appear in my images and therefore I ignored it as in this sample at a high ISO. My attitude is that, if I cannot see it any practical way, it does not matter, in spite of theory.

Of course theory is always tucked away in my mind.


12800 ISO 5D ll

--
Charles Darwin: "ignorance more frequently begets confidence than does knowledge."
tony

 

[bobn2]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

For the Canon 1D and 5D the reduction was both large and also insignificant. Large because the read noise reduced by a factor of around 8 as the ISO was increased from 100 to 3200. Insignificant because the fixed pattern noise was so bad (and objectionable) that it rendered the whole question of the read noise level moot.

In pretty much all modern cameras the fixed pattern noise has been reduced to a level which makes it effectively not a problem. These cameras have a nice even random noise pattern which isn't visually objectionable.

The noise reduction is certainly there for most cameras, but isn't really a big issue. It's talked about mostly just as a counterpoint when someone says wrongly that increasing ISO increases noise.

Here are four Canon 5 cameras:



These are all at 100 ISO and visually not much different, even though the read noise is quite different.

According to Bill Claff's figures at 100 ISO the read noise (per pixel) is 27 e- for the 5D, 13 e- for the Mk IV, 11,5 for the Ds and 4.1 for the R5 (misleading because canon has applied involuntary NR to get there). Now let's look at 3200 ISO



We can see that the 5D is worse, and this tallies with Bill's figures, which are 4 e- for the 5D, 2 e- for the MK IV and Ds and 1.5 for the R5 (no NR this time). So, why can't we see all the extra noise at 100 ISO? The simple reason is that it's all below the black level. The 5D can collect about 50,000 electrons per pixel. Given a normal headroom that would mean that the white level for a shot exposed for 100 ISO would be around 28,000. That puts the read noise level about 1000 times lower, for an effective DR of 10 stops. That's quite enough to fit a full tonal scale from white to black. If, however, you'd set 100 ISO but exposed for 3200, that sets the white level at 875 electrons. Now the read noise is only 5 stops below the white, so is going to be above a normally set black level.

Also, from Bill's curves you'll find that the improvement above 3200 is pretty marginal in most cameras.

This clarifies some of my concerns.

In terms of "noise measurement", is the Photoshop standard deviation of a consistent size image patch an appropriate measure of noise as apparently done in D Cox original post? It is, in my estimation, at least a good comparative measure.

When we talk about 'noise' in photography we're actually talking about signal to noise ratio, (or actually the noise to signal ratio, if we want it to scale so bigger = more noise) so Don should really have taken both the SD and the mean and divided the former by the latter. Then there is the problem that he's measuring noise in a processed file, not the raw. This makes sense if you're interested in what the noise looks like at the end, but not if you're trying to test out a minor variance in the raw. Raw Digger would have been a better tool to use than PS. So it's not a silly way of doing it, but for what he was trying to do introduces confounding factors in terms of possible differences in the processing.

Although fixed pattern noise on the Canon 5D ll and 5D was easy to demonstrate, it did not actually appear in my images and therefore I ignored it as in this sample at a high ISO. My attitude is that, if I cannot see it any practical way, it does not matter, in spite of theory.

Absolutely. Why do you think that Canon has sold so many cameras in spite of offering lower DR than other brands. In usual usage in good light the noise is way below the black level and you just can't see it in a normal processed photo.

The problem with both noise and DR is that people like quantified figures of merit, even if the don't know what they mean. 14 stops DR versus 12 stops allows people to say that it's better, even if the difference doesn't show in normal practice.

Of course theory is always tucked away in my mind.


12800 ISO 5D ll

The Mk II is much better than the original. It has the same sensor as the 1Ds IIi, which in that camera ($7k !) was limited to 3200 ISO although it had the best low-light sensor that Canon made at the time.

--
Is it always wrong
for one to have the hots for
Comrade Kim Yo Jong?

 

[FingerPainter]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant.

I believe the following two photos were posted here by bobn2 as part of a series several years ago. They were taken RAW in the same light at identical exposures but different ISO settings. The image lightnesses were normalized in development, and no NR was applied.







I can see a difference, Can you? To me, the ISO 1600 shot is visibly less noisy than the ISO 100 shot.

You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It seems the advantage from increasing ISO decreases as ISO setting increases. The improvement from increasing ISO from 100 to 400 seems usually to be greater than the improvement from ISO 400 to 1600, for instance, The advantage over ISO 3200 is often minimal or even on-existant. Of course, these break points vary between cameras.

If ISO increases at high ISO levels are implemented with digital adjustments rather than increased gain, there will be no advantage to noise performance whatsoever. Perhaps I'm misinterpreting them, but Bill Claff's PDR charts seem to me to indicate that increases to ISO setting above 1600 on the 1DII and 5d, and above ISO 3200 on the IDIII and above 6400 on the 5DII were done differently than lower increases. On teh charts fro some other cameras, Bill specifically indicates scaling being used to implement a high ISO increase.

Another factor that may come into play is that past a certain increase in analog gain, there is no advantage to be gained at the low bit count end due to the unamplified signal being totally buried below the noise floor, to the extent that nothing is gained by applying more gain to these very low pixel values. But ask an engineer. like bobn2. He'll know a lot more than my idle speculation.

It would be nice to see a practical example of the noise reduction for real world shooting.

Well I hope the two images above are an adequate illustration.

Am I misunderstanding or missing a salient point ?

Well you talked about effects at high ISOs, while the example is at relatively low ISOs. Bill Claff's shadow improvement charts show improvement tapering off as ISO increases.

 

[bobn2]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant.

I believe the following two photos were posted here by bobn2 as part of a series several years ago. They were taken RAW in the same light at identical exposures but different ISO settings. The image lightnesses were normalized in development, and no NR was applied.





I can see a difference, Can you? To me, the ISO 1600 shot is visibly less noisy than the ISO 100 shot.

This was done as a series of exposure workshops that I did for some photo clubs (anyone still want them, I'm happy to oblige). I constructed it to allow the participants to conduct experiments with their own cameras in real time. So in this, I didn't take the photos, it was a photographer using his own camera. We set up a scene, they set the camera to 6400 in M mode, metered for that, took the shot and the changed the ISO only and repeated the shot. Then they gave me the card and I ran a script on my computer to process them. It didn't equalise lightness, what it actually did was multiply the raw numbers by an appropriate amount to equalise them (if you like, applying 'digital gain' - all done in floating point so there were no rounding errors and nowhere for additional noise to creep in) - then the raws went through exactly the same processing - so that no-one could claim that there was something uncontrolled in the processing. Then we projected the results on the big screen.

I think that doing it themselves made it very easy for the participants to grasp what was happening, and to separate out the concerns of exposure and ISO.

--
Is it always wrong
for one to have the hots for
Comrade Kim Yo Jong?

 

[ThrillaMozilla]

When we talk about 'noise' in photography we're actually talking about signal to noise ratio, (or actually the noise to signal ratio, if we want it to scale so bigger = more noise) so Don should really have taken both the SD and the mean and divided the former by the latter.

It should be that simple if the zero point were correct, but I don't think he measured raw values. As far as I can tell, he measured on a scale of 0 to 255 (jpeg), which is a profoundly nonlinear scale. So it doesn't work at all.

Then there is the problem that he's measuring noise in a processed file, not the raw. This makes sense if you're interested in what the noise looks like at the end, but not if you're trying to test out a minor variance in the raw. Raw Digger would have been a better tool to use than PS. So it's not a silly way of doing it, but for what he was trying to do introduces confounding factors in terms of possible differences in the processing.

To do it from his images, one also has to measure the slope, which he didn't do.

 

[ThrillaMozilla]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant.

I believe the following two photos were posted here by bobn2 as part of a series several years ago. They were taken RAW in the same light at identical exposures but different ISO settings. The image lightnesses were normalized in development, and no NR was applied.

Yes, this is exactly what happens. It's much easier to understand from the picture than from the numbers. DPR does this with every camera review. They call it an ISO-invariance test, and the effect is easy to see.

 

[ThrillaMozilla]

In terms of "noise measurement", is the Photoshop standard deviation of a consistent size image patch an appropriate measure of noise as apparently done in D Cox original post? It is, in my estimation, at least a good comparative measure.

There would have been a very different result if he had measured a dark part of the image instead of the brightest part. See my first post, which I have revised yet again.

 

[bobn2]

When we talk about 'noise' in photography we're actually talking about signal to noise ratio, (or actually the noise to signal ratio, if we want it to scale so bigger = more noise) so Don should really have taken both the SD and the mean and divided the former by the latter.

It should be that simple if the zero point were correct, but I don't think he measured raw values. As far as I can tell, he measured on a scale of 0 to 255 (jpeg), which is a profoundly nonlinear scale. So it doesn't work at all.

Indeed, I forgot all about that.

Then there is the problem that he's measuring noise in a processed file, not the raw. This makes sense if you're interested in what the noise looks like at the end, but not if you're trying to test out a minor variance in the raw. Raw Digger would have been a better tool to use than PS. So it's not a silly way of doing it, but for what he was trying to do introduces confounding factors in terms of possible differences in the processing.

To do it from his images, one also has to measure the slope, which he didn't do.

Absolutely.

 

[tony field]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

For the Canon 1D and 5D the reduction was both large and also insignificant. Large because the read noise reduced by a factor of around 8 as the ISO was increased from 100 to 3200. Insignificant because the fixed pattern noise was so bad (and objectionable) that it rendered the whole question of the read noise level moot.

In pretty much all modern cameras the fixed pattern noise has been reduced to a level which makes it effectively not a problem. These cameras have a nice even random noise pattern which isn't visually objectionable.

The noise reduction is certainly there for most cameras, but isn't really a big issue. It's talked about mostly just as a counterpoint when someone says wrongly that increasing ISO increases noise.

Here are four Canon 5 cameras:



These are all at 100 ISO and visually not much different, even though the read noise is quite different.

According to Bill Claff's figures at 100 ISO the read noise (per pixel) is 27 e- for the 5D, 13 e- for the Mk IV, 11,5 for the Ds and 4.1 for the R5 (misleading because canon has applied involuntary NR to get there). Now let's look at 3200 ISO



We can see that the 5D is worse, and this tallies with Bill's figures, which are 4 e- for the 5D, 2 e- for the MK IV and Ds and 1.5 for the R5 (no NR this time). So, why can't we see all the extra noise at 100 ISO? The simple reason is that it's all below the black level. The 5D can collect about 50,000 electrons per pixel. Given a normal headroom that would mean that the white level for a shot exposed for 100 ISO would be around 28,000. That puts the read noise level about 1000 times lower, for an effective DR of 10 stops. That's quite enough to fit a full tonal scale from white to black. If, however, you'd set 100 ISO but exposed for 3200, that sets the white level at 875 electrons. Now the read noise is only 5 stops below the white, so is going to be above a normally set black level.

Also, from Bill's curves you'll find that the improvement above 3200 is pretty marginal in most cameras.

This clarifies some of my concerns.

In terms of "noise measurement", is the Photoshop standard deviation of a consistent size image patch an appropriate measure of noise as apparently done in D Cox original post? It is, in my estimation, at least a good comparative measure.

When we talk about 'noise' in photography we're actually talking about signal to noise ratio, (or actually the noise to signal ratio, if we want it to scale so bigger = more noise) so Don should really have taken both the SD and the mean and divided the former by the latter.

I don't think he tried to look at noise as a numeric precise metric. It was a reasonable comparison for higher or lower generalizations as in "is the noise going up or down". This is indicated above 1270 ISO and is the point where his camera performs in an "isoless" manner as indicated in Bill's Shadow improvement vs ISO chart.

I see no photographic use to explore the concept of noise associated with extreme ISO variations and constant exposure for legitimate photography.

My interest was only for high ISO noise variation when shooting in quite low light.

Then there is the problem that he's measuring noise in a processed file, not the raw. This makes sense if you're interested in what the noise looks like at the end, but not if you're trying to test out a minor variance in the raw. Raw Digger would have been a better tool to use than PS. So it's not a silly way of doing it, but for what he was trying to do introduces confounding factors in terms of possible differences in the processing.

Although fixed pattern noise on the Canon 5D ll and 5D was easy to demonstrate, it did not actually appear in my images and therefore I ignored it as in this sample at a high ISO. My attitude is that, if I cannot see it any practical way, it does not matter, in spite of theory.

Absolutely. Why do you think that Canon has sold so many cameras in spite of offering lower DR than other brands. In usual usage in good light the noise is way below the black level and you just can't see it in a normal processed photo.

The problem with both noise and DR is that people like quantified figures of merit, even if the don't know what they mean. 14 stops DR versus 12 stops allows people to say that it's better, even if the difference doesn't show in normal practice.

--
Charles Darwin: "ignorance more frequently begets confidence than does knowledge."
tony

 

[bobn2]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

For the Canon 1D and 5D the reduction was both large and also insignificant. Large because the read noise reduced by a factor of around 8 as the ISO was increased from 100 to 3200. Insignificant because the fixed pattern noise was so bad (and objectionable) that it rendered the whole question of the read noise level moot.

In pretty much all modern cameras the fixed pattern noise has been reduced to a level which makes it effectively not a problem. These cameras have a nice even random noise pattern which isn't visually objectionable.

The noise reduction is certainly there for most cameras, but isn't really a big issue. It's talked about mostly just as a counterpoint when someone says wrongly that increasing ISO increases noise.

Here are four Canon 5 cameras:



These are all at 100 ISO and visually not much different, even though the read noise is quite different.

According to Bill Claff's figures at 100 ISO the read noise (per pixel) is 27 e- for the 5D, 13 e- for the Mk IV, 11,5 for the Ds and 4.1 for the R5 (misleading because canon has applied involuntary NR to get there). Now let's look at 3200 ISO



We can see that the 5D is worse, and this tallies with Bill's figures, which are 4 e- for the 5D, 2 e- for the MK IV and Ds and 1.5 for the R5 (no NR this time). So, why can't we see all the extra noise at 100 ISO? The simple reason is that it's all below the black level. The 5D can collect about 50,000 electrons per pixel. Given a normal headroom that would mean that the white level for a shot exposed for 100 ISO would be around 28,000. That puts the read noise level about 1000 times lower, for an effective DR of 10 stops. That's quite enough to fit a full tonal scale from white to black. If, however, you'd set 100 ISO but exposed for 3200, that sets the white level at 875 electrons. Now the read noise is only 5 stops below the white, so is going to be above a normally set black level.

Also, from Bill's curves you'll find that the improvement above 3200 is pretty marginal in most cameras.

This clarifies some of my concerns.

In terms of "noise measurement", is the Photoshop standard deviation of a consistent size image patch an appropriate measure of noise as apparently done in D Cox original post? It is, in my estimation, at least a good comparative measure.

When we talk about 'noise' in photography we're actually talking about signal to noise ratio, (or actually the noise to signal ratio, if we want it to scale so bigger = more noise) so Don should really have taken both the SD and the mean and divided the former by the latter.

I don't think he tried to look at noise as a numeric precise metric. It was a reasonable comparison for higher or lower generalizations as in "is the noise going up or down". This is indicated above 1270 ISO and is the point where his camera performs in an "isoless" manner as indicated in Bill's Shadow improvement vs ISO chart.

I see no photographic use to explore the concept of noise associated with extreme ISO variations and constant exposure for legitimate photography.

My interest was only for high ISO noise variation when shooting in quite low light.

He was trying to test the hypothesis that noise reduces as ISO increases. It's a minor effect and experimental technique needs to be good to test it out. As you say, in most usage it's not relevant. It generally only gets mentioned when the old myth that ISO increases noise is trotted out.

--
Is it always wrong
for one to have the hots for
Comrade Kim Yo Jong?

 

[ThrillaMozilla]

I don't think he tried to look at noise as a numeric precise metric. It was a reasonable comparison for higher or lower generalizations as in "is the noise going up or down".

I think he failed, mainly because he didn't look in the right place, but also because he didn't do it right. You have to be careful how you do experiments. Had he not compared only the brightest part of the image, and had he posted pictures instead of questionable numbers, he might have done better.

This is indicated above 1270 ISO and is the point where his camera performs in an "isoless" manner as indicated in Bill's Shadow improvement vs ISO chart.

...and as I showed you. Claff's shadow improvement chart is just another view of the input-referred noise.

I see no photographic use to explore the concept of noise associated with extreme ISO variations and constant exposure for legitimate photography.

My interest was only for high ISO noise variation when shooting in quite low light.

Then it is probably not useful for you. If you had a Nikon D7200, for example, or if you worked with somewhat higher exposures, you might find shooting at unusually low ISO settings to be more useful. In any case, this method of shooting is basically a kludge that sometimes works to overcome a quirk in camera designs.

 

[Chris Noble]

We set up a scene, they set the camera to 6400 in M mode, metered for that, took the shot and the changed the ISO only and repeated the shot. Then they gave me the card and I ran a script on my computer to process them. It didn't equalise lightness, what it actually did was multiply the raw numbers by an appropriate amount to equalise them (if you like, applying 'digital gain' - all done in floating point so there were no rounding errors and nowhere for additional noise to creep in) - then the raws went through exactly the same processing - so that no-one could claim that there was something uncontrolled in the processing.

This is an useful demonstration of the beneficial effect of raising analog (pre-Raw) ISO to pull the shadows above the bottom of the ADC conversion range. What counts in the end is how the photo looks.

 

[fferreres]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

For the Canon 1D and 5D the reduction was both large and also insignificant. Large because the read noise reduced by a factor of around 8 as the ISO was increased from 100 to 3200. Insignificant because the fixed pattern noise was so bad (and objectionable) that it rendered the whole question of the read noise level moot.

In pretty much all modern cameras the fixed pattern noise has been reduced to a level which makes it effectively not a problem. These cameras have a nice even random noise pattern which isn't visually objectionable.

The noise reduction is certainly there for most cameras, but isn't really a big issue. It's talked about mostly just as a counterpoint when someone says wrongly that increasing ISO increases noise.

Here are four Canon 5 cameras:



These are all at 100 ISO and visually not much different, even though the read noise is quite different.

According to Bill Claff's figures at 100 ISO the read noise (per pixel) is 27 e- for the 5D, 13 e- for the Mk IV, 11,5 for the Ds and 4.1 for the R5 (misleading because canon has applied involuntary NR to get there). Now let's look at 3200 ISO



We can see that the 5D is worse, and this tallies with Bill's figures, which are 4 e- for the 5D, 2 e- for the MK IV and Ds and 1.5 for the R5 (no NR this time). So, why can't we see all the extra noise at 100 ISO? The simple reason is that it's all below the black level. The 5D can collect about 50,000 electrons per pixel. Given a normal headroom that would mean that the white level for a shot exposed for 100 ISO would be around 28,000. That puts the read noise level about 1000 times lower, for an effective DR of 10 stops. That's quite enough to fit a full tonal scale from white to black. If, however, you'd set 100 ISO but exposed for 3200, that sets the white level at 875 electrons. Now the read noise is only 5 stops below the white, so is going to be above a normally set black level.

Also, from Bill's curves you'll find that the improvement above 3200 is pretty marginal in most cameras.

I find the 5D (classic) not too bad in terms of noise. You make a great point of why we don't see noise at ISO 100. A good profiling of it's noise could get it into a very low threshold where 3200 would look really well. If one had the resources, could be easily trained via machine learning with a wide set of photos at every ISO, so it'd not only dramatically reduce the noise by also correct the color.

 

[John Sheehy]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

How far is far?

If there is any difference in input-referred read noise at various ISO settings, whether or not you see it depends on exposure, subject matter, and light color, as well as display brightness of the shadow areas. When one of the three color channels has very weak sensor capture (think halogen for blue or deep forest shade for red), read noise becomes much stronger (relatively) in that channel, and is more likely to be visible.

 

[tony field]

There is an often repeated declaration that high ISO reduces noise. This is caused by the signal processing when the sensor image is processed and conditioned for the ADC. Although the reduction does happen, I have never seen this to be visually significant. You can infer this in D Cox thread a week ago With his SNR test using Photoshop to indicate the STD deviation of the test patches. Indeed a small improvement of SNR above ISO 1270 does happen.

I shot Canon 1D and 5D series extensively for low light performing arts between iso 3200 and ISO 12800 but have never seen a significant improvement of SNR. A number of years ago when this was first discussed, I did the appropriate test and, with pixel peeping, the noise reduction does exist but not to any degree to be of an advantage when shooting theatre or dance.

It would be nice to see a practical example of the noise reduction for real world shooting.

Am I misunderstanding or missing a salient point ?

How far is far?

If there is any difference in input-referred read noise at various ISO settings, whether or not you see it depends on exposure, subject matter, and light color, as well as display brightness of the shadow areas. When one of the three color channels has very weak sensor capture (think halogen for blue or deep forest shade for red), read noise becomes much stronger (relatively) in that channel, and is more likely to be visible.

Indeed those attributes can affect the appearance of noise. Would this be a strong difference in normal shooting at ISO 1600 to 12800 of normal scenes for performing arts? Does noise decrease with increasing ISO? As far as I can determine, a fractional decrease does happen but not to any photographically useful extent