I know, I know, it can be caused by various factors, like high ISO, poor lighting conditions, long exposure time, heat. (not that I know why the last two cause noise)
But where does noise REALLY come from? What is noise? What does it have to do with photons and photosites?
Noise can come from the gain stage - where electricity is used to control the “ISO” of the sensor. Or it can come from the encoding stage - the codec turning raw bytes into a digital file.
https://en.wikipedia.org/wiki/Image_noise
Does a better example of explaining it in depth than I can.
That’s a great resource, here is another that has more visuals on sensors and how they work. https://graphics.stanford.edu/courses/cs448a-10/sensors-noise-14jan10-opt.pdf
Edit: Page 36 is where signal to noise ratio is directly addressed.
- EM interference from circuits on the sensor
- heat and IR radiation from the sensor
- imperfect ADC circuits
- wave-like behaviour of light due to quantum effects
With modern cameras the noise is mostly inherent randomness in the number of photons that hit that particular pixel. If you want to find the ”real” root cause, it’s simply the random behavior of the light source which sends out photons at some rate (aka brightness) where the time and direction of each individual photon is random.
The more photons that pixel collects, the more that randomness is averaged out and thus the less noisy the image is.
This article series on DPReview is decent.
Noise is just miniscule variations caused by myriad of factors - heat, humidity, different amount of photons, properties of different electronic components, natural backround radiation etc. etc. etc…
And no… high ISO doesn’t cause noise, it just amplifies it.
Life, uh, finds a way
And no… high ISO doesn’t cause noise, it just amplifies it.
No it doesn’t.
ISO setting is a lightness parameter. Adjusting it ajusts JPG lightnesss. This is very different from amplifying noise.
Regarding raw files and the underlaying information, the sensor tries to capture - using a high ISO does not amplify noise. Instead it on typical camera slightly reduces the little noise that the analogue to digital convenrsion adds, and some cases also reduces the noise that the pixel adds.
I appreciate your input, but instead of misinterpreting Wikipedia articles you should learn how electronics actually work. :)
This was a rabbit hole I went down recently that was quite illuminating.
Sensors have a native ISO and any extended (higher or lower) are simply the sensor amplifying or decreasing the signal.
Sensors have a native ISO
No they do not. ISO is an output format (e.g. JPG) metric defined by ISO 12232 standard. It has nothing to do with image sensors.
Most cameras do change the operational parameters of the image sensor when the ISO setting is changed, typically the PGA (programmable gain amplifier) setting is changed and the signal is amplified the more the higher the camera’s ISO setting is.
and any extended (higher or lower) are simply the sensor amplifying or decreasing the signal.
That is not right.
Typically the ISO 100 (sometimes something else) setting on a camera is such that the image sensor is run at the lowest PGA amplification setting. Anything above that and the amplification is increased (and/or digital multiplication is used in software). The “extended lo” settings typically operate the sensor at the very same setting the ISO 100 does, just change the metering and processing of the data.
high ISO doesn’t cause noise, it just amplifies it.
I think of it like trying to listen to a really quiet video or audio file on headphones or speakers. You can keep turning it up until you can hear what sounds are on that file. But you will also be turning up the background white noise
I put ants in your camera
This camera needs to be…at least three times this size!
One source: There is always noise from the electrons in the wires, moving about because they have thermal energy. If your signal is weak and you turn up the gain (aka ISO in a digital camera) then it can amplify this noise up along with the signal. The thermal noise can be reduced by cooling the electronics. At absolute zero there would be no thermal noise…
If you’re talking film the answer is film has literal grain because it’s physical.
If you’re talking digital the answer is sensors use gain to increase and decrease sensitivity. When you increase the gain you start having cross-talk as you overload individual pixels and the detection kinda bleed across them.
That’s a very broad simplified explanation as I understand it.
If you’re talking film the answer is film has literal grain because it’s physical
Grain is only one part of the noise function. The other is the random nature of light itself.
If you’re talking digital the answer is sensors use gain to increase and decrease sensitivity
Nope. We can of course define “sensitivity” is multiple ways, but typically in this context the image sensors have one sensitivity (though different for each wavelength of light).
When you increase the gain you start having cross-talk as you overload individual pixels and the detection kinda bleed across them.
No, this is plan wrong.
Here’s what happens typically:
- Lens draws a noisy image (because of photon shot noise - light is noisy, and this noise dominates outside of very light starved places)
- Pixels capture this image - at this stage pixels add such a tiny amount of noise to the signal that is usually irrelevant.
- Signal moves to PGA (programmable gain amplifier) - here the analogue signal is amplified according to selected settings (typically the ISO setting controls this)
- Signal moves to ADC (analogue to digital converter) where it is digitized - this process is slightly noisy, thus some noise is added to the signal
- A raw file is written or the camera processes a JPG and writes it
The reason why there is step 3 is in step 4 noise. If you increase the signal before ADC, the noise that the ADC adds becomes less relevant in comparison. The drawback for amplification is that the signal may “burn”, or the ADC operational range is exceeded.
A fun experiment is that you take two shots with identical exposure settings: one at ISO 100, the other at ISO 6400, both in raw format, and then process them in Lightroom or other such raw processor to the same lightness. For practically all today’s cameras the ISO 6400 will appear cleaner.
Quantum stuff. Electrons and photons are discrete packets of energy.
It’s said that disgruntled painters found a grimoire in the 1800s and cast a curse upon photography for all of time
Satisfied French noises
The source is quantum in nature. Every now and then an electron will break through and light up the pixel. Not much can be done about it happening, but software can “fix” things sometimes.
It also comes from my wife when I whip out the DSLR at a particularly interesting location while on vacation and say “I’ll just be a minute…”
The kids! That’s where the noise comes from. I struggle to find anything that can create more noise than them 😖
What is noise?
Typically noise is deviation from the expected value. Expected value can be measured by taking multiple samples and the more samples you get, the more precise the expected value will be. In practise expected value is the average of the samples. Thus noise is deviation from that. Signal to noise ratio (SNR) tells you how much noise there is.
But where does noise REALLY come from?
Light itself is noisy - light particles, photons, hit the image plane randomly, following poisson distribution. Thus the image that the lens draws is in itself noisy. The more light particles you collect, the larger the SNR will be.
Almost all the noise in the raw-file data comes from randomness of light.
Image sensor adds a tiny little amount to this and it’s irrelevant unless very little light is present.