High ISO and noise with correct exposure

[Brocken]

Nikon's Hi Iso noise reduction systems operates by:

1. Generating an image of the noise alone at the chosen ISO value by taking a picture without opening the shutter.

2. Subtracting this noise image from the chosen picture so the noise pixels are cancelled out.

Electronic noise is highly dependent on temperature. Higher temperature mean more noise, and lower temperature less noise.
This dependence on temperature explains why we can't make a single noise image and apply it to all our pictures: not all will have been taken at the same temperature.

[Manfred M]

Ali as per my previous reply, correct exposure and noise have nothing to do with each other. Noise is related to the ISO level you are shooting at and correct exposure is related to the aperture and shutter speed used to capture the image at that ISO level.

Simply said, a correctly exposed image at a low ISO setting will have less noise than a correctly exposed image at a high ISO setting. You get a similar issue when using film; lower sensitivity film tends to have finer grain than more sensitive high speed film. The grain is a function of sensitivity, and correct exposures are possible (and desireable) with both high-speed and low-speed film.

When you use flash, or for that matter any form of lighting, where there are dark areas like the shadows in the image, that is where you are going to see noise. The best way to reduce sensor noise is to shoot at the lowest ISO you can get away with.

[Alis]

Add on to my other reply

The correctly exposed back ground is only likely to happen if you use fill in flash or a variation of it. Another way of putting that is that a correct exposure of the background in your shot might over expose the subject. What I suspect has happened is that the flash has been used to set the exposure of the subject and that the actual exposure time is insufficient to record the background correctly. It may have helped a bit but the main lighting has come from the flash - background is further away so there is less light intensity on it. If you have a problem understanding that googling "flash guide numbers" should help. These are used to set manual flash exposures purely based on distance and aperture. and of course iso.

-

Thanks, John. There was no problem understanding this part. This is in fact what I have done for this picture. I explained since you mentioned in your post that the foreground may get overexposed.

[Manfred M]

On higher end video cameras we have the same function that Christopher describes and it is called Automatic Black Balance (ABB) and you run that every so often and it does exactly what he describes.

Sensors used to Astrophotography are cooled for precisely the reason Christopher statesl to reduce the noise component in the electronic components caused by heat.

[Alis]

Ali as per my previous reply, correct exposure and noise have nothing to do with each other. Noise is related to the ISO level you are shooting at and correct exposure is related to the aperture and shutter speed used to capture the image at that ISO level.

Simply said, a correctly exposed image at a low ISO setting will have less noise than a correctly exposed image at a high ISO setting. You get a similar issue when using film; lower sensitivity film tends to have finer grain than more sensitive high speed film. The grain is a function of sensitivity, and correct exposures are possible (and desireable) with both high-speed and low-speed film.

When you use flash, or for that matter any form of lighting, where there are dark areas like the shadows in the image, that is where you are going to see noise. The best way to reduce sensor noise is to shoot at the lowest ISO you can get away with.

Thanks, Manfred. This is clear now!

[Alis]

Nikon's Hi Iso noise reduction systems operates by:

1. Generating an image of the noise alone at the chosen ISO value by taking a picture without opening the shutter.

2. Subtracting this noise image from the chosen picture so the noise pixels are cancelled out.

Electronic noise is highly dependent on temperature. Higher temperature mean more noise, and lower temperature less noise.
This dependence on temperature explains why we can't make a single noise image and apply it to all our pictures: not all will have been taken at the same temperature.

Very interesting, Christopher!

I thought the problem with noise is that it is random. So, not two images will be the same. That I thought was the idea behind image stacking. That because of that randomness, when you stack pictures of the same scene, the noisy areas of one image get covered by the noiseless areas of others.

[Alis]

BTW, so all of you guys know, nothing fascinates me more than a digital image sensor. The only exception is: how little stuff you can find about it. They never go deeper than certain level or they are too technical with a lot of calculus and algebra that I give up after reading the first paragraph!

[Brocken]

Very interesting, Christopher!

I thought the problem with noise is that it is random. So, not two images will be the same. That I thought was the idea behind image stacking. That because of that randomness, when you stack pictures of the same scene, the noisy areas of one image get covered by the noiseless areas of others.

I am sorry I did not make myself clearer. Dark frame subtraction, as this technique is known, is especially effective at cancelling fixed pattern noise to which digital sensors are prone, especially with long exposures where pixel charge leakage becomes a problem, and/or where there are 'hot' pixels, (pixels with a high dark current, when no light is falling on the sensor) and 'stuck' pixels which don't react to light at all.

As for noise itself, it is true that noise due to the thermal agitation of electrons in a conductor is random, but other types of noise are dependent on one or more variables. For example, flicker noise is related to frequency, while quantization noise is caused by the process of converting the analogue pixel voltage into the discrete steps needed by digital circuits. Think of having to draw a circle using only vertical and horizontal lines in a great quantity of tiny steps. The greater the number of these tiny steps, the better we will approximate a true circle, but a true circle it can never be, because there are no straight lines in a circle. The difference between the true curve and these steps is itself a source of noise.

[xpatUSA]

So, my theoretical question is: If I am a single pixel and I am correctly exposed, do I generate a lot of noise?

Am I being dumb and you guys already answered my question or no?

Cheers!

As a single pixel, you would generate noise all the time, but only one kind of noise - often called "dark noise" but known in the electronics world as "leakage current" when talking about reverse-biased diodes. This current is always flowing if the diode (in our case, photodiode) has some volts across it. The current is random in nature and has a slightly different average value in each of your fellow-pixels. This current increases or decreases by a factor of two approximately every 9 deg C. Because the current is always flowing, the longer the exposure the more noise will appear in your well (diode capacitance).

People sometimes talk about signal-to-noise ratio (SNR) as if it had a single value applicable to a camera's electronics under all conditions. But dark parts of an image only received a few photons at the sensor so that means the ratio of electrons produced by the light to electrons produced by noise can be quite low and that ratio is the SNR of that area in your image. So now you can see that a dark area with a long exposure time will look more noisy than a bright area with that same long exposure time.

You're not being dumb, ya just need "more input" (like No. 5) re: image sensor noise, photon noise, Johnson noise, Poisson noise, conversion noise, etc, ad nauseam. Go forth and Googlify, my son!!

[revi]

Nikon's Hi Iso noise reduction systems operates by:

1. Generating an image of the noise alone at the chosen ISO value by taking a picture without opening the shutter.

2. Subtracting this noise image from the chosen picture so the noise pixels are cancelled out.

Electronic noise is highly dependent on temperature. Higher temperature mean more noise, and lower temperature less noise.
This dependence on temperature explains why we can't make a single noise image and apply it to all our pictures: not all will have been taken at the same temperature.

As noise is random (by definition), subtracting a black image will not completely cancel the electronic noise (unfortunately, or we wouldn't have the high-ISO noise problems ). But it will diminish the noise (theoretically by a factor of 1.4 iirc).

And you will get good correction for any systematic deviations in your sensor (as these are at fixed spots, and are also amplified for high ISO). An example of such systematic deviations are the hot pixels in long exposures. Correction for these works the same way, by taking a second identical exposure w/o opening the shutter (or putting the lens cap in place) and subtracting it from the picture. As the hot pixels in long exposures are a bit less sensitive to temperature, you can use one black image for a series of normal images with the same exposure time.

[Alis]

So now you can see that a dark area with a long exposure time will look more noisy than a bright area with that same long exposure time.

Great, Ted! This is what I was suspecting.

And when I see someone talking fluently as you do about digital sensors and he is from Texas, I assume you have something to do with Texas Instruments? Or may be another company like that?

Thanks anyway!

[ajohnw]

Very interesting, Christopher!

I thought the problem with noise is that it is random. So, not two images will be the same. That I thought was the idea behind image stacking. That because of that randomness, when you stack pictures of the same scene, the noisy areas of one image get covered by the noiseless areas of others.

That's perfectly correct with that style of stacking it reduces if it's truly random. Makes me think of Murphy's and that piece of grass's law and exposure time but the post would be too long.
The confusion come about because there are several sources of noise and different types or styles of it. They might be called, white or pink for for instance. If you want a better understanding astro pages might provide it via google and terms like dark frame, flat frame. There is some info here found by searching ccd noise. Probably similar information is available for cmos sensors as well but problems are similar. Noise is an inherent problem in electronics that may or not matter. It's fundimental.

http://www.mssl.ucl.ac.uk/www_detect...rkcurrent.html

-

[xpatUSA]

And when I see someone talking fluently as you do about digital sensors and he is from Texas, I assume you have something to do with Texas Instruments? Or may be another company like that?

Born, went to school and worked for most of my life in England. Been in Texas for over 25 years and now I cain't hardly speak good English ;-)

[Colin Southern]

I think we need to be careful with definitions here Ali - when you say "correct exposure" I interpret that as potentially having 2 meanings;

1. A darkly exposed background may be "correct" because it - wait for it - is actually dark (eg photographer has made an artistic choice to have it that way). In that case the data will be closer to the noise floor - there WILL be more noise associated with that portion of the exposure (ie lower signal to noise ratio), but it's not a problem because you won't be able to see it. It's only if you subsequently try to reveal the detail in the dark areas that both the signal and the noise will be amplified and thus possibly revealing noise.

2. And of course there's the usual definition of correct exposure (eg tones looking "normal").

In broad & real-world terms High ISO modes don't cause VISIBLE noise per se - it's the under-exposure that - when corrected - produces the visible noise. If there's no need to raise the levels of a dark are then noise is in essence a moot point; regardless of whether it's there or not doesn't matter if it can't be seen. Bit like "does a tree falling along in the forest where nobody can hear it still make a sound" (short answer is "who cares"!).

[revi]

And as to WHY it is better to use a higher iso than to 'correct' under-exposure in PP:
Sensor values are analog values, meaning that any value within a given range is possible*. Pixel values are integer values, so there's only a limited number of values available.
Somewhere in the process, the sensor values will get rounded to an integer to get pixel values.
If we use high ISO, the sensor value is multiplied by a factor, and then rounded.

So if, in a zone of flat colour, we have one pixel at 5.45 and another at 5.54 (raw reading at 100 ISO), then multiply to get 800 ISO, those pixels will end up with values of 43.6 and 44.3. Rounded, we get 44 for both => same colour, so no visible noise. (Sensor values were not identical => there is noise, as we have a flat colour area)

Now, we save as 100 ISO. Our pixels will then be stored as values of 5 (5.45 rounded) and 6 (5.54 rounded). To get the same colour in our final image, we will have to multiply those values by 8 in PP. So we end up with final values of 40 and 48 (still average of 44). But now we have a difference between the pixels of 8! That's visible noise...

Of course, this is a contrived example, and a rather extreme correction. But the idea still holds for more realistic values: correction in PP will aggravate any rounding errors that could have been avoided by correctly exposing at a high ISO, rather than underexposing at a lower ISO. And this will lead to exaggerated noise in the final image