Lenses and ISO noise

L

LCD

Member
Hello,

I know performance at high ISO depends mostly, if not solely on your camera.... but how much, if any, does your glass effect ISO noise? In other words, when I upgrade my glass for my 60D, will it decrease the artifacts I get, specifically from 800 ISO and up? I hardly ever shoot anything that high but I'm curious.

Thanks
-LCD-
 
LCD said:
Hello,

I know performance at high ISO depends mostly, if not solely on your camera.... but how much, if any, does your glass effect ISO noise? In other words, when I upgrade my glass for my 60D, will it decrease the artifacts I get, specifically from 800 ISO and up? I hardly ever shoot anything that high but I'm curious.

Thanks
-LCD-
Click to expand...

Noise is caused by the sensor. What happens is that the sensor does many calculations based on photons hitting the sensor. Think of it like a poll. A poll is most accurate when you have more people take part in it. Noise is simply too few photons hitting the sensor for the equations to work. The camera calculations then error and create a wrong pixel, which will stand out from its neighbors, thus the spec of noise. Now imagine that happens all over at once. That is noise. Noise are simply pixels that are incorrect because the camera lacked the information to properly represent what should be there. At least that is how I think of it.

The lens won't have anything directly to do with noise, however if it's a slow lens, it is helping starve the sensor causing the noise.
 
Ryan Patrick O'Hara said:
Noise is caused by the sensor. What happens is that the sensor does many calculations based on photons hitting the sensor. Think of it like a poll. A poll is most accurate when you have more people take part in it. Noise is simply too few photons hitting the sensor for the equations to work. The camera calculations then error and create a wrong pixel, which will stand out from its neighbors, thus the spec of noise. Now imagine that happens all over at once. That is noise. Noise are simply pixels that are incorrect because the camera lacked the information to properly represent what should be there. At least that is how I think of it.

The lens won't have anything directly to do with noise, however if it's a slow lens, it is helping starve the sensor causing the noise.
Click to expand...

Thanks for the break down... well said!
 
Ryan Patrick O'Hara said:
Noise is caused by the sensor. What happens is that the sensor does many calculations based on photons hitting the sensor. Think of it like a poll. A poll is most accurate when you have more people take part in it. Noise is simply too few photons hitting the sensor for the equations to work. The camera calculations then error and create a wrong pixel, which will stand out from its neighbors, thus the spec of noise. Now imagine that happens all over at once. That is noise. Noise are simply pixels that are incorrect because the camera lacked the information to properly represent what should be there. At least that is how I think of it.

The lens won't have anything directly to do with noise, however if it's a slow lens, it is helping starve the sensor causing the noise.
Click to expand...

In all the explanations I've heard of what noise is and how it's created (and I've heard tons... both good and bad), that's the first I've ever heard that one.
Not that I'm saying you're wrong. But I haven't heard it before, so if I wanted to read more into your explanation... where would I go?

edit: or I'm misunderstanding you? The way you make it sound, noise is an incorrect reading of signals by the sensor (essentially). The way I understand it, noise is a CORRECT reading of signals, but it's reading the "wrong" signal (essentially).
 
That is an interesting point of reference in regard to digital video noise. What ever works. It is my understanding that digital video sensors still have a pure analog output, X amount of photons excite the struck pixels to output a directly proportional electrical signal; more photons, more electricity, less photons less electricity. The noise is caused by the increase in the gain that is applied to the sensor output to bring it up to a useable threshold; that is "Signal Noise". Sensor design and output processing design, both analog and digital, help to decrease the noise.

To the OP, Like Ryan says, the lens' effect on noise is dependent on its speed.
 
sonofaresiii said:
In all the explanations I've heard of what noise is and how it's created (and I've heard tons... both good and bad), that's the first I've ever heard that one.
Not that I'm saying you're wrong. But I haven't heard it before, so if I wanted to read more into your explanation... where would I go?

edit: or I'm misunderstanding you? The way you make it sound, noise is an incorrect reading of signals by the sensor (essentially). The way I understand it, noise is a CORRECT reading of signals, but it's reading the "wrong" signal (essentially).
Click to expand...



It is much more complicated than I made it sound... but basically, where there is very little light, the sensor does not have enough information for it's algorithms which deturmine what the pixel will look like when amplified. When there is not enough information, some pixels are incorrectly 'painted' and thus they stand out from their correct neighbors. These are the specs we see that make up noise. Blue channel noise, a Red1 favorite for years, was caused by a starved blue channel, thus the noise in that channel.

Just saying, it's like a poll, the results become skewed possibly erroring when there is not enough information... this is noise. No>?
 
Last edited:
How David stated it is the same thing. The sensor is forced to use amplification (gain) of the signal to get an image, but when you don't have a healthy amount of information to amplify the results can become inaccurate. I suppose that is a good way of incorporating his statement.
 
sonofaresiii said:
edit: or I'm misunderstanding you? The way you make it sound, noise is an incorrect reading of signals by the sensor (essentially). The way I understand it, noise is a CORRECT reading of signals, but it's reading the "wrong" signal (essentially).
Click to expand...

The circuit is reading the 'correct value' from the sensing cell... The cell itself has had only a 'few photons' hit it because of the low light and so it reads out with a more 'random' value. For example, on frame time 1, it reads out that 20 photons hit it... on frame time two, as luck would have it, 40 photons hit it... hence the pixel on the second frame would be 2x the pixel value on the first.

This is all simplified of course but that's the idea...

I think of the process like putting a sheet of paper out in a hard rain. Within a few seconds the sheet of paper is soaked. If the rain is light, only a few drops are on the paper, they will appear at 'random' places on the paper every time you set it out in the rain.
 
j1clark@ucsd.edu said:
The circuit is reading the 'correct value' from the sensing cell... The cell itself has had only a 'few photons' hit it because of the low light and so it reads out with a more 'random' value. For example, on frame time 1, it reads out that 20 photons hit it... on frame time two, as luck would have it, 40 photons hit it... hence the pixel on the second frame would be 2x the pixel value on the first.

This is all simplified of course but that's the idea...

I think of the process like putting a sheet of paper out in a hard rain. Within a few seconds the sheet of paper is soaked. If the rain is light, only a few drops are on the paper, they will appear at 'random' places on the paper every time you set it out in the rain.
Click to expand...

interesting analogy. If that were true, wouldn't we always have noise in our shadows or dim scenes? I mean, if it were from random photons? I figured it was incorrect algorithms on the sensor side, and not photons on the real-world side.

Interesting. Waiting for some coding engineer to come in and school us all.
 
Well, try this explanation:
A warm sensor generates its own noise. As it gets warmer, it generates even more noise. When you turn up the gain, you're increasing the voltage to the sensor, which causes it to get warmer, thus you get more noise.
Years ago, I was assisting at the Mount Laguna astronomical observatory. To read dim stars, the sensor must be kept very cold, otherwise the data will be "lost in the weeds" of the sensor noise. So one of my duties was to pack dry ice into the box which surrounds the sensor.

High ISO = high voltage to sensor = high temperature of sensor = more noise :cry:

-- Ken

P.S. -- So I guess the solution is to turn the ISO down ... which will require using more lights ... which will generate more heat ... <sigh>
 
Last edited:
Ryan Patrick O'Hara said:
interesting analogy. If that were true, wouldn't we always have noise in our shadows or dim scenes? I mean, if it were from random photons? I figured it was incorrect algorithms on the sensor side, and not photons on the real-world side.

Interesting. Waiting for some coding engineer to come in and school us all.
Click to expand...

That has what had me scratching my head many times; why, under very similar lighting levels, exposure and camera settings, do some shots have more noise than others. I should find some shoots like that and really look at them on the scopes.

Ken Hull said:
Well, try this explanation:
A warm sensor generates its own noise. As it gets warmer, it generates even more noise. When you turn up the gain, you're increasing the voltage to the sensor, which causes it to get warmer, thus you get more noise.
Years ago, I was assisting at the Mount Laguna astronomical observatory. To read dim stars, the sensor must be kept very cold, otherwise the data will be "lost in the weeds" of the sensor noise. So one of my duties was to pack dry ice into the box which surrounds the sensor.

High ISO = high voltage to sensor = high temperature of sensor = more noise :cry:

-- Ken

P.S. -- So I guess the solution is to turn the ISO down ... which will require using more lights ... which will generate more heat ... <sigh>
Click to expand...

Yeah, video with the DSLRs really brought up the heat issue with some cameras. With a T2i I remember seeing increasing noise on longer shoots when I knew the sensor was heating up. Maybe we will see super sensors that are water cooled!!! :2vrolijk_08:
 
Ken Hull said:
Well, try this explanation:
A warm sensor generates its own noise. As it gets warmer, it generates even more noise. When you turn up the gain, you're increasing the voltage to the sensor, which causes it to get warmer, thus you get more noise.
Years ago, I was assisting at the Mount Laguna astronomical observatory. To read dim stars, the sensor must be kept very cold, otherwise the data will be "lost in the weeds" of the sensor noise. So one of my duties was to pack dry ice into the box which surrounds the sensor.

High ISO = high voltage to sensor = high temperature of sensor = more noise :cry:

-- Ken

P.S. -- So I guess the solution is to turn the ISO down ... which will require using more lights ... which will generate more heat ... <sigh>
Click to expand...

I signed up to assist a grad student at the observatories on Mt. Laguna... and every assigned day/night for a year it was overcast... I determined that was a message from the gods that they didn't want me spying on them...

Here's a web page that discusses noise and various contributions for digital imaging for 'stills'. There are references to various Canon models, so I think it is more applicable than some other discussions that one may run across from the 'astronomy' imaging side.

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/

The article breaks up 'noise' into several components/contributors.


Photon shot noise
Read noise
Pattern noise
Thermal noise
Pixel response non-uniformity
Quantization error

Especially interesting is the discussion "S/N ratio vs. exposure, and Dynamic Range" where the article uses the Canon 40D as an example, where the 'theoretical' max 14 bits of the digitizer, and a 14 stop representation... is reduced to about 7 stops because the actual voltage values don't reach the 14 bit max, the noise of the sensor from all sources, limits the lowest value, and the resulting difference between noise and max, yields a 11 stop range... but then some number of values at the low end do not yield 'useful' images, which adds (or subtracts) up to a net of about 7 stops of effective DR...

Since higher ISO values have 'more noise', I think that probably explains to some extent why there is a reduction in Dynamic Range for higher ISO values.
 
j1clark@ucsd.edu said:
I signed up to assist a grad student at the observatories on Mt. Laguna... and every assigned day/night for a year it was overcast... I determined that was a message from the gods that they didn't want me spying on them...

Here's a web page that discusses noise and various contributions for digital imaging for 'stills'. There are references to various Canon models, so I think it is more applicable than some other discussions that one may run across from the 'astronomy' imaging side.

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/

The article breaks up 'noise' into several components/contributors.


Photon shot noise
Read noise
Pattern noise
Thermal noise
Pixel response non-uniformity
Quantization error

Especially interesting is the discussion "S/N ratio vs. exposure, and Dynamic Range" where the article uses the Canon 40D as an example, where the 'theoretical' max 14 bits of the digitizer, and a 14 stop representation... is reduced to about 7 stops because the actual voltage values don't reach the 14 bit max, the noise of the sensor from all sources, limits the lowest value, and the resulting difference between noise and max, yields a 11 stop range... but then some number of values at the low end do not yield 'useful' images, which adds (or subtracts) up to a net of about 7 stops of effective DR...

Since higher ISO values have 'more noise', I think that probably explains to some extent why there is a reduction in Dynamic Range for higher ISO values.
Click to expand...

I assumed we were gonna get a slice of complicated pie.

In cinematography, we've done tests, and found that over exposing slightly (by using a ISO higher than factory) actually increases range, while using very low ISO's, like 200 or 100 typically gives a bit less. I don't know why, but that were the results we got. :/

Don't know why.
 
Ryan Patrick O'Hara said:
I assumed we were gonna get a slice of complicated pie.

In cinematography, we've done tests, and found that over exposing slightly (by using a ISO higher than factory) actually increases range, while using very low ISO's, like 200 or 100 typically gives a bit less. I don't know why, but that were the results we got. :/

Don't know why.
Click to expand...

I suspect there is a difference between 'lab' calibration and analysis and 'what looks good'. I wish the manufacturers would just give out the specs, what they designed for, etc. and let people who want to know these sorts of details use them. As it is they seem to bury this sort of info, unless one is shooting on a higher end camera. (It is of course an assumption that this sort of info is available for RED and ARRI and other similar classes of cameras.)

In a sense as it is, one has to 'reverse' engineer what the camera will do... beyond 'point and shoot'.

But then one had to do that in the Film film days as well, although for many people there had been 50 or more years of 'knowledge' developed about what could be done with film.
 
You must log in or register to reply here.
Back
Top