-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
ISO Invariance: an experiment
- Thread starter Tom Axford
- Start date
mamallama
Forum Pro
Who said anything about increase or decrease?? I said it's high from the start and stays high throughout.
confused circle
Senior Member
- Messages
- 1,932
- Reaction score
- 1,554
I don't think we did? Changing the iso doesn't change the exposure unless the camera is in a mode where it automatically selects shutter or aperture. But in such a mode, a lower iso should lead to a higher exposure, since more light will need to reach the sensor to give the same jpeg brightness as a shot at a higher iso.
I thought we weren't taking about ettr at or near base iso.
I'm probably thinking about the ETTR thread when I say that.
It's nothing specific. Did I give that impression? I don't recall.
confused circle
Senior Member
- Messages
- 1,932
- Reaction score
- 1,554
I probably just got hung up on the iso 6400 part
FingerPainter
Forum Pro
- Messages
- 12,593
- Solutions
- 37
- Reaction score
- 13,643
I think you are confusing the amount of signal coming from the scene with the amount of signal captured by the sensor.
A scene can be very well lit, but if your shutter is fast enough and your aperture narrow enough, only a small signal is captured. The shot noise of the captured image depends on the amount of signal captured, not the amount of signal in the scene.
mamallama
Forum Pro
True, but capturing enough signal for the test was no problem.. The test scene was bright with no dark shadows thus it was bright with no dynamic range challenges. The camera sensor adds shot noise. ISO invariant sensors are low noise and add little noise which sets them apart for regular sensors. So both test pictures started out with high S/N ratio. That was my point.
John Sheehy
Forum Pro
Any sensor that uses only one analog gain and digitization is completely ISO-invariant, noise-wise. Close to that, there are some cameras that have so much pre-gain read noise relative to post-gain read noise, that they may as well only have one gain, even if they have multiple ones. Most of these are older or special-purpose cameras.
Ultimately, we are moving towards total ISO invariance, when sensors just count photons, but along the way, as long as we have read noise, the ratio of post-gain to pre-gain read noise, as well as the implementation of conversion gain, has a lot to say to prevent ISO setting absolute noise invariance (which is what "ISO invariance" is intended to mean, I think.
Besides the amount of noise, there is uncharted spatially-correlated post-gain read noise, converter/metadata problems, hardware clippings and quantizations, etc, which make ISO-invariance sometimes harder to realize than the charts might suggest, if we really push things.
The most ISO-invariant cameras, counter-intuitive to many people's perception, are older ones with the most noise. As pre-gain read noise has been dropping, post-gain read noise gets revealed again as a major read noise source at low ISOs, and needs to be reduced further.
Yes. For example, here's a photo of a scene that is well lit but, through control of aperture, was rendered dark with minimal signal captured during the exposure - f/16, 1/20, ISO 100:
Here's the same scene at f/16, 1/20, ISO 3200:
And a 100% crop of the second image...
I did make a -0.5 stop EC adjustment in Lightroom.
Followed by a 100% crop of the first image after a +3.5 stop EC adjustment in Lightroom:
Clearly, both exposures (f/16, 1/20) capture minimal signal. The dark appearance of the first image at base ISO demonstrates this fact. What helps the exposures succeed is the relatively shallow dynamic range within the scene. Lightening the initial exposure by +3.5 stops (whether in-camera using ISO or out-of-camera using Lightroom's exposure comp) produces a pleasingly light image with minimal shading. If a significant portion of the scene had been shaded or obstructed somehow from the ambient light of the scene, even less signal - and a correspondingly lower signal-to-noise ratio - would have been captured in those areas. Lightening (by ISO or in LR) would have rendered the noise more obvious in those areas.
--
Bill Ferris Photography
Flagstaff, AZ
riveredger
Senior Member
It does not, it adds read, etc., noise.
They could very well be high noise.
A few things are being demonstrated by this set of examples.Yes. For example, here's a photo of a scene that is well lit but, through control of aperture, was rendered dark with minimal signal captured during the exposure - f/16, 1/20, ISO 100:
Here's the same scene at f/16, 1/20, ISO 3200:
And a 100% crop of the second image...
I did make a -0.5 stop EC adjustment in Lightroom.
Followed by a 100% crop of the first image after a +3.5 stop EC adjustment in Lightroom:
Clearly, both exposures (f/16, 1/20) capture minimal signal. The dark appearance of the first image at base ISO demonstrates this fact. What helps the exposures succeed is the relatively shallow dynamic range within the scene. Lightening the initial exposure by +3.5 stops (whether in-camera using ISO or out-of-camera using Lightroom's exposure comp) produces a pleasingly light image with minimal shading. If a significant portion of the scene had been shaded or obstructed somehow from the ambient light of the scene, even less signal - and a correspondingly lower signal-to-noise ratio - would have been captured in those areas. Lightening (by ISO or in LR) would have rendered the noise more obvious in those areas.
FingerPainter
Forum Pro
- Messages
- 12,593
- Solutions
- 37
- Reaction score
- 13,643
What do you mean by "enough signal for the test"? The point of the test seems to have been to capture a low enough signal that there would be visible shot noise and to see what effect different ISO settings had on total noise.
The test image that didn't need boosting in development was taken at settings of
1/1000 f/8 ISO 6400
That indicates the scene was EV 10, which is half way between day;light and a normally lit interior room at night. That is not part particularly bright. The fact that there were no shadows indicates a lack of strong direct light. It does not indicate bright lighting.
Bright light produces scenes with a greater dynamic range than low light.
No, shot noise in is the captured light. The camera adds several other sorts of noise, which can be grouped together under the term "camera-added noise".
Not necessarily. And it is cameras that are ISO Invariant, not sensors. ISO Invariant cameras trend to be ones that add little noise after the variable gain stage. They may add plenty of noise before that point in the processing chain.
Both pictures started out with about 5 stops more noise than they would have in direct sunlight. That means their SNR was was only about 17.7% of what it would have been in daylight at base ISO.
Your point seems to be without foundation.
mamallama
Forum Pro
Call it what you will, but the noise is from the sensor. It's a low noise sensor and the that is what sets them apart from regular sensors.
It's a low noise sensor.
No, it is a constant (with respect to ISO) noise sensor, low or high.
Iliah Borg
Forum Pro
The word "shot" here is for a reason different from that it is a photographic shot. It is after shot towers.
Photon shot noise is because of the particle nature of light.
Sensor has dark current shot noise, which is negligibly small.
mamallama
Forum Pro
What do you think is the source of the noise?? Whatever comes through the lens is the signal. It is just the light from the scene you are trying to capture.
Not so. There are two types of gain involved in the total process: analog gain and digital gain. Analog gain occurs in first stage in the sensor, digital gain occurs after the A/D process and, in this case, during the RAW processing. The advantage of the ISO invariant sensor is that this analog gain is achieved by a very low noise amplifier so the S/N is reasonably high even if the camera exposure is below that normal sensors. After the first amplification the signal is then high enough that the subsequent amplifiers do not have to be low noise. And after the signal is digitized, digital amplification adds no additional noise. The resulting S/N of the output (picture) is basically set by the first stage in the ISO invariant sensor which has a low noise base.
This whole process is exactly analogous to the amplification chain of a radio or radar receiver. Only the first stage needs to be low noise because after the first stage, the signal is much larger than any noise that would be added in subsequent stages.
What does this have to do with the test?
The test only shows the principle of the ISO invariant sensor works. No one is disputing that.
Tom Axford
Forum Pro
Light contains its own noise due to the quantum nature of light. Light consists of photons and the photons arrive randomly. This randomness appears as noise in the image. Even a perfect camera (which introduces no noise itself) will still give noisy images because of the quantum nature of light.
So, what comes through the lens is not just the signal, it contains noise as well.
mamallama
Forum Pro
That's at another level. Make no mistake, the noise we are talking about in photography is optical noise from the process itself and is not the quantum noise nature of light. In the analogy of a radio or radar receiver, noise that comes through the antenna would be considered interference noise. This, from the dpr article on 'ISO-invariance' is the noise we are talking about that contributes to the dynamic range of the sensor.
"Sony sensors tend to show very high Raw dynamic range due to low noise characteristics, but we wanted to take a closer look at exactly how little noise the camera itself contributes to images".
https://www.dpreview.com/articles/7450523388/sony-alpha-7r-ii-real-world-iso-invariance-study
Last edited:
FingerPainter
Forum Pro
- Messages
- 12,593
- Solutions
- 37
- Reaction score
- 13,643
The source of shot noise is random variation in the light that is due to the quantum nature of the process which emits photons from a light source. The timing and wavelengh of released photons is the product a bounded random process, which results in random variation in light colour and intensity, but that has a particular distribution.
Whatever comes through the lens is a signal and its associated shot noise. It is not the same as the signal which is captured by the cameras because a significant portion of the light that comes through the lens is not captured, for a number of reasons, including:
- Some of the light coming through the lens does not fall on the sensor,
- If the sensor has a CFA, (most do), then only some of the photons, those whose wavelength is in the range passed by the element of the CFA they strike, get through to the sensor.
- Of those photons that get through the CFA, only some manage to release an electron when they strike the sensor.
For OP's images, the shot noise in the captured signal is still more significant than the camera-added noise.
No, it is not just a signal. The light has noise in it.
What exactly is "not so"?
The digital manipulation which may occur is not properly termed "gain". When I talk about the "variable gain stage" I am talking about analog gain. Most cameras add more noise before the analog variable gain stage than after it.
What do you mean by "first stage"?
Digital "gain" is a red herring. I am not talking about it.
No. ISO invariance generally has noting to do with noise being added by an amplifier. In those cases where noise is added by an amplifier, if it is proportional to the amount of amplification, treat it as noise added before amplification, and if it is independent of the amount of amplification, treat it as noise added after amplification.
Amplifiers used in digital cameras have little to no effect on SNR. The amplify both the signal and the noise by the same amount so the SNR after gain is roughly the same as before gain.
True, but after digitization is not what I'm talking about.
You are omitting the effect of noise added after the gain stage but before digitization.
In fact, on some cameras, the noise added at or after the variable analog gain stage is still significantly large enough that it plays a significant role in the final total SNR. When it is not amplified, but the signal and any noise present before the variable gain stage are are both amplified, the total SNR rises as the amplification increases. Generally, cameras in which post-gain-stage noise is insignificant are the ones which approach ISO invariancy.
It has to do with your claim that the SNR was high.
Similar threads
Keyboard shortcuts
- f
- Forum