Doug Kerr
Well-known member
ABSTRACT AND INTRODUCTION
We generally consider that, for a given camera model, noise performance (for example, the signal-to-noise ratio on a scene area of low luminance) degrades with increasing setting of ISO sensitivity.
But in fact the ISO sensitivity setting only has that effect indirectly.
Actually:
• for a scene element in general, its signal-to-noise ratio in the image is determined by the photometric exposure the element receives on the sensor, regardless of ISO sensitivity.
• for a given element in a specific scene, its signal-to-noise ratio in the image is determined by the (photographic) exposure used for the shot, regardless of ISO sensitivity.
The reason that the ISO sensitivity setting appears to influence noise performance is that, for a certain scene, when we set a certain ISO sensitivity:
• It influences the (photographic) exposure the metering system causes the camera to use (for metered exposure)
or
• It influences the (photographic) exposure we choose in pursuit of a particular "exposure strategy"
Failure to recognize this can lead us to misunderstand what various "exposure techniques" might do.
***************
EXAMPLES
Lets look at some examples of how this works. We will start with a certain ISO sensitivity, and we take a shot using the exposure chosen by the camera's metering system (Shot A). We somehow measure the noise performance, in terms of "signal to noise ratio", for a dark object in the image.
Next (for shot B), we will leave that ISO setting in effect and take a second shot with half the exposure (manually set, for example). Because of the lesser photographic exposure, the photometric exposure for any scene "element" is now half what it was in shot A. And we will find that the signal-to-noise ratio for our scene element of interest is now worse.
Assuming that we do not apply any "push" during development of the image, than scene element is now at a lower point on the image tonal scale, which we might not want. But these examples are to illustrate the working of the principle being discussed, not to suggest a photographic practice!
Next (for shot C), we set the ISO sensitivity to twice what it was for shot A, and we take a shot with the same photographic exposure used for shot B. Since, from shot B, we have changed neither the luminance of scene objects nor the photographic exposure, the photometric exposure for any scene element is the same as in shot B. And in fact, we will find that the signal-to-noise ratio for our scene element of interest is essentially the same as in shot B.
Here, our scene element is at essentially the position in the tonal scale of the image as in shot A.
So far, we have been working either with teh camera JPG output or with external development of the camera raw data (but with no "push").
For shot D, we return to the setup for shot B, but apply a one-stop "push" during development. Again, the photometric exposure is half what it was in the "normally metered" shot, shot A. But the "push" in development places the scene element of interest in the same place in the image tonal scale as in shot A.
And we will find that the signal-to-noise ratio for our scene element of interest is essentially the same as in shot B.
(I have to keep saying "essentially" since for various reasons these relationships do not hold exactly in practice.)
Now, for shot E, we will use the same topographical exposure as in shot A (which was "normally metered" but set the ISO sensitivity to a value two times as much as for shot A.
Since the scene is the same, and the photographic exposure the same, the photometric exposure will be the same.
But very possibly, in the "developed" image (again assume "standard" development, perhaps to the camera JPG output), the highlights may be "blown out". THus this is probablu not something we would actually want to do. But if we examine teh image, we will find that, for the scene element of interest, the signal-to-noise ratio is essentially the same as for shot A.
WHAT THIS ISN'T
It is not the object of this note to suggest exposure strategies for various situations. It is rather to make sure that, as we consider such, we keep our eye "on the right ball" with regard to how ISO sensitivity affects the noise result.
SUMMARY
Noise performance (in terms of signal-to-noise ratio) essentially depends on the photometric exposure for the scene element of interest.
Even more practically:
For a given scene element in a specific scene (having a given luminance), the signal to noise ratio essentially depends only on the (photographic) exposure.
This is not at all to say that the choice of ISO sensitivity is of no consequence. It is very important. It is a factor that we must use to fulfill other objectives, such as avoidance of "highlight blowout". But the relationship in bold above will always be with us as we do that.
We generally consider that, for a given camera model, noise performance (for example, the signal-to-noise ratio on a scene area of low luminance) degrades with increasing setting of ISO sensitivity.
But in fact the ISO sensitivity setting only has that effect indirectly.
Actually:
• for a scene element in general, its signal-to-noise ratio in the image is determined by the photometric exposure the element receives on the sensor, regardless of ISO sensitivity.
• for a given element in a specific scene, its signal-to-noise ratio in the image is determined by the (photographic) exposure used for the shot, regardless of ISO sensitivity.
The reason that the ISO sensitivity setting appears to influence noise performance is that, for a certain scene, when we set a certain ISO sensitivity:
• It influences the (photographic) exposure the metering system causes the camera to use (for metered exposure)
or
• It influences the (photographic) exposure we choose in pursuit of a particular "exposure strategy"
Failure to recognize this can lead us to misunderstand what various "exposure techniques" might do.
***************
EXAMPLES
Lets look at some examples of how this works. We will start with a certain ISO sensitivity, and we take a shot using the exposure chosen by the camera's metering system (Shot A). We somehow measure the noise performance, in terms of "signal to noise ratio", for a dark object in the image.
Next (for shot B), we will leave that ISO setting in effect and take a second shot with half the exposure (manually set, for example). Because of the lesser photographic exposure, the photometric exposure for any scene "element" is now half what it was in shot A. And we will find that the signal-to-noise ratio for our scene element of interest is now worse.
Assuming that we do not apply any "push" during development of the image, than scene element is now at a lower point on the image tonal scale, which we might not want. But these examples are to illustrate the working of the principle being discussed, not to suggest a photographic practice!
Next (for shot C), we set the ISO sensitivity to twice what it was for shot A, and we take a shot with the same photographic exposure used for shot B. Since, from shot B, we have changed neither the luminance of scene objects nor the photographic exposure, the photometric exposure for any scene element is the same as in shot B. And in fact, we will find that the signal-to-noise ratio for our scene element of interest is essentially the same as in shot B.
Here, our scene element is at essentially the position in the tonal scale of the image as in shot A.
So far, we have been working either with teh camera JPG output or with external development of the camera raw data (but with no "push").
For shot D, we return to the setup for shot B, but apply a one-stop "push" during development. Again, the photometric exposure is half what it was in the "normally metered" shot, shot A. But the "push" in development places the scene element of interest in the same place in the image tonal scale as in shot A.
And we will find that the signal-to-noise ratio for our scene element of interest is essentially the same as in shot B.
(I have to keep saying "essentially" since for various reasons these relationships do not hold exactly in practice.)
Now, for shot E, we will use the same topographical exposure as in shot A (which was "normally metered" but set the ISO sensitivity to a value two times as much as for shot A.
Since the scene is the same, and the photographic exposure the same, the photometric exposure will be the same.
But very possibly, in the "developed" image (again assume "standard" development, perhaps to the camera JPG output), the highlights may be "blown out". THus this is probablu not something we would actually want to do. But if we examine teh image, we will find that, for the scene element of interest, the signal-to-noise ratio is essentially the same as for shot A.
WHAT THIS ISN'T
It is not the object of this note to suggest exposure strategies for various situations. It is rather to make sure that, as we consider such, we keep our eye "on the right ball" with regard to how ISO sensitivity affects the noise result.
SUMMARY
Noise performance (in terms of signal-to-noise ratio) essentially depends on the photometric exposure for the scene element of interest.
Even more practically:
For a given scene element in a specific scene (having a given luminance), the signal to noise ratio essentially depends only on the (photographic) exposure.
This is not at all to say that the choice of ISO sensitivity is of no consequence. It is very important. It is a factor that we must use to fulfill other objectives, such as avoidance of "highlight blowout". But the relationship in bold above will always be with us as we do that.