A Question about ISO and Noise.

[shahmatt]

So my question is: Is ISO directly proportional to noise?

More specifically, what I want to know is if its truly proportional, or does the rate of noise-increase change as ISO is raised?

If the noise level at an ISO of 800 is say X, would the noise level of ISO1600 be 2X?

Anyone know?

[Rooz]

So my question is: Is ISO directly proportional to noise?

More specifically, what I want to know is if its truly proportional, or does the rate of noise-increase change as ISO is raised?

If the noise level at an ISO of 800 is say X, would the noise level of ISO1600 be 2X?

Anyone know?

its almost impossible to answer this question cos there are so many factors in the equation.

[TheWengler]

I don't think it's linear (that's a guess), but I agree with Rooz that it's more than just ISO causing noise in an image. The best way to improve your ISO is to nail your exposure. If you underexpose and fix it in photoshop then that will make the noise worse.

[cdifoto]

So my question is: Is ISO directly proportional to noise?

More specifically, what I want to know is if its truly proportional, or does the rate of noise-increase change as ISO is raised?

If the noise level at an ISO of 800 is say X, would the noise level of ISO1600 be 2X?

Anyone know?
It will vary from camera to camera because each has its own processing Al-Gore-Rhythms going on. So all you can do is crank it up and see. If you can't tell, it doesn't matter.

[D Thompson]

its almost impossible to answer this question cos there are so many factors in the equation.

+1 what Rooz said. A underexposed shot at ISO-pick one will have more noise than a well exposed shot at ISO-pick-one.

[Beowulff]

Yep... what Rooz said.

One can only say with any certainty that (obviously) the higher the ISO, the more the noise.

:)

[shahmatt]

Well, I think my question needs a little clarifying.

Basically what I want to know is if noise can be predicted.

Lets say I have an image of ISO 1600 (RAW with no post processing or noise reduction).

I believe there is a measurement for noise, I can't remember the unit though. But lets say this image has x noise units.

Now from that, can I predict that an image of ISO 3200 will have 2x noise units, and an image of ISO 800 will have x/2 noise units?

This would be assuming other variables are the same!

[K1W1]

Are you assuming the all pictures will be of identical subjects under identical lighting conditions?

[shahmatt]

Well yeah. Its more of a theory question really!

[cdifoto]

Well, I think my question needs a little clarifying.

Basically what I want to know is if noise can be predicted.

Lets say I have an image of ISO 1600 (RAW with no post processing or noise reduction).

I believe there is a measurement for noise, I can't remember the unit though. But lets say this image has x noise units.

Now from that, can I predict that an image of ISO 3200 will have 2x noise units, and an image of ISO 800 will have x/2 noise units?

This would be assuming other variables are the same!
You'd have to get some sorta geeky sensor analysis program to run the numbers. That would determine your camera's behavior, and then you can predict based on that.

As far as the real world non-theory goes, I know that on my particular bodies, a higher ISO means more noise in my images but higher ISO also means faster shutter speeds at the same aperture. I don't care if it's 1.73x, 2x, or E=mc², I know it's more and will adjust accordingly.

[Rooz]

I don't care if it's 1.73x, 2x, or E=mc², I know it's more and will adjust accordingly.

yepp, yepp. spot on.

[AdamW]

You'd have to get some sorta geeky sensor analysis program to run the numbers. That would determine your camera's behavior, and then you can predict based on that.

As far as the real world non-theory goes, I know that on my particular bodies, a higher ISO means more noise in my images but higher ISO also means faster shutter speeds at the same aperture. I don't care if it's 1.73x, 2x, or E=mc², I know it's more and will adjust accordingly.

Also, noise is affected by things like the temperature of the sensor, so the atmospheric conditions would have to be the same.

[Mark_48]

Plug the known variables into the equations in the image shown below. This may render an approximate baseline for the thermal noise generated by the sensor. With that baseline, consider the non-linear gain structure of the sensor sensitivitity as the ISO is proportionally increased. From a correlation of thermally generated noise and electron gain provided by the sensor, an estimate of the resulting degree of photonic image noise can be approximated.

Disclaimer: I don't have a clue what I just said, but doesn't it sound almost convincing? :D

A read here may provide a more realistic overview on noise and ISO..
http://en.wikipedia.org/wiki/Image_noise

[Beowulff]

.....I believe there is a measurement for noise, I can't remember the unit though.

I'm guessing that you're confusing aural "noise" — which is any non-harmonious or discordant group of sounds — with image noise. Or you may be referring to harmonic distortion (which can be measured).

But lets say this image has x noise units.You can't actually quantify image "noise" in any meaningful sense, other than subjectively. Even if you could assign some base level for a logarithmic scale, it wouldn't work (see below).

Now from that, can I predict that an image of ISO 3200 will have 2x noise units, and an image of ISO 800 will have x/2 noise units?In a word — no.

The multiple technical factors involved with image noise are far too complex to isolate and evaluate. Most higher ISO "noise" is caused by the sensor's photo-diodes being unable to uniformly measure the photons falling upon them, and to consistently replicate that measurement for a given set of conditions. The variations in this measurement are also dependent on the intensity of the light source, and to make matters worse, not all photo-diodes display the same degree of variation above or below a theoretical "average" for a given intensity.

I guess an analogy to answer the question you're asking is to say that yes, it's definitely gonna rain tomorrow, but whether we get 20mm or 200mm is completely unknown until after the event. Same with the noise in the pic you snap.

Cheers :)

[Mark_48]

I'm guessing that you're confusing aural "noise" — which is any non-harmonious or discordant group of sounds — with image noise. Or you may be referring to harmonic distortion (which can be measured).

You can't actually quantify image "noise" in any meaningful sense, other than subjectively. Even if you could assign some base level for a logarithmic scale, it wouldn't work (see below).

In a word — no.

The multiple technical factors involved with image noise are far too complex to isolate and evaluate. Most higher ISO "noise" is caused by the sensor's photo-diodes being unable to uniformly measure the photons falling upon them, and to consistently replicate that measurement for a given set of conditions. The variations in this measurement are also dependent on the intensity of the light source, and to make matters worse, not all photo-diodes display the same degree of variation above or below a theoretical "average" for a given intensity.

I guess an analogy to answer the question you're asking is to say that yes, it's definitely gonna rain tomorrow, but whether we get 20mm or 200mm is completely unknown until after the event. Same with the noise in the pic you snap.

Cheers :)
This might be close.....
http://en.wikipedia.org/wiki/Signal_to_noise_ratio_(image_processing)

[shahmatt]

This might be close.....
http://en.wikipedia.org/wiki/Signal_to_noise_ratio_(image_processing)

This one seems to have an answer. So noise follows a log curve of sorts, which is probably why ISO is typically doubled in cameras: 100, 200, 400, 800, 1600 etc...

Just to be clear, I know that noise increases with ISO. There is no debate here.

My question is whether noise increases in exact proportion to ISO. If that's the case we can make a prediction on what the level of noise will be for different ISO's without having to actually take the shot and compare by eye.

So if I knew the noise at a certain ISO, I can predict the noise for other ISOs accurately. This is provided that noise can be quantified in dB from a sample image.

Anyway, I think this is enough technical talk for now. Hehe! Thanks for the input guys.

[cdifoto]

This one seems to have an answer. So noise follows a log curve of sorts, which is probably why ISO is typically doubled in cameras: 100, 200, 400, 800, 1600 etc...

Just to be clear, I know that noise increases with ISO. There is no debate here.

My question is whether noise increases in exact proportion to ISO. If that's the case we can make a prediction on what the level of noise will be for different ISO's without having to actually take the shot and compare by eye.

So if I knew the noise at a certain ISO, I can predict the noise for other ISOs accurately. This is provided that noise can be quantified in dB from a sample image.

Anyway, I think this is enough technical talk for now. Hehe! Thanks for the input guys.
You don't need to measure it or compare images. You just need to shoot your camera and get to know it.