Why Are My Photos Noisy?

Well, I am not actually asking the question in this thread. This thread is about my understanding, in general layman terms, of what noise is, what causes it and how to minimise it.

What is Noise?

There are two sources of noise in an image.

1. Shot Noise.
   This is usually the vast majority of the noise in an image. It is due to the random way in which light strikes the sensor.

2. Read Noise.
   Read noise is caused by the internal workings of the camera (electronics, heat etc) during the processing of the data recorded at the sensor to its storage in a raw or jpeg file on the camera's memory card.

What Makes Noise Visible in an Image?

The more light that hits the sensor while the shutter is open, the higher will be the Signal to Noise Ratio (SNR) in the data. I think of SNR as the ratio of good data to bad data. The higher the SNR, the less visible will be the noise. Therefore, it is a low SNR in the data that will make noise visible in the image.
Consequently, the larger the exposure*, the higher will be the SNR resulting in the visible noise being minimum, if at all.

How To Minimise Visible Noise

Basically, you need to maximise the exposure* within your DOF and motion blur or other artistic constraints without clipping important highlights. In other words, you need to get as much light as possible onto the sensor while the shutter is open within your DOF and motion blur constraints without clipping important highlights.

There are a few ways you can do that. My preferred methods depending on the situation are:

1. Set the largest aperture (smallest f-number) that will give my DOF and the slowest shutter speed that will meet my motion blur constraints without clipping important highlights and then Auto ISO. This can be done in Manual Mode + Auto ISO. If the camera sets base ISO and important highlights are still being clipped you then obviously need to compromise on dof and/or motion blur. If there is no motion in the scene then the decision is easy and you can use a faster shutter speed until important highlights are no longer clipped.

2. Set aperture priority mode and then the largest aperture (smallest f-number) that will give my DOF. I then set the minimum shutter speed setting to be the minimum shutter speed that will meet my motion blur constraints and then Auto ISO. In this configuration, if at base ISO the camera wants to set a slower shutter speed than my minimum, the camera will set the minimum shutter speed and raise ISO to suit. If at base ISO the camera determines there will be highlight clipping at the minimum shutter speed, then the camera maintains base ISO and sets a faster shutter speed to avoid highlight clipping.

3. Add light to the scene or wait for better light. Obviously that is normally much easier said than done unless you are in a studio.

All 3 methods above will maximise the exposure*, within my DOF and blur constraints without clipping important highlights, which will maximise the SNR resulting in minimising the visible noise.

Note though, since the camera's histogram is based on the jpeg data the camera generates from the raw data, the histogram is not necessarily a 100% true indicator of when the raw data is being clipped but in my experience the histogram is a very close approximation for the raw data.

* exposure - amount of light striking the sensor per unit area while the shutter is open.
The 3 parameters that affect the exposure are aperture, shutter speed and the lighting of the scene.

There are more, but maybe saying "two main sources, two fundamental limits" could cover it.

Traditionally, read noise is considered independent of temperature of the sensor. It only covers the noise generated during the readout process, where the voltage induced by this charge is measured, pre-ADC (analogue-to-digital converter).

How does exposure affect shot noise?
Below "e-" stands for "electron".
Each visible light photon interacting with the silicone photodiode in the pixel creates a single electron, and, in summary, signal(e-) is close to be proportional to the amount of light the pixel collected.
Numerically,
Shot noise (e-) = sqrt (signal(e-)); where sqrt stands for square root
SNR = signal(e-) / shot noise (e-) = signal(e-) / sqrt (signal(e-)) = sqrt (signal(e-))

Thus, if we have 100 photoelectrons accumulated in a pixel, the shot noise-caused SNR is 10; but if we have 10,000 electrons, the SNR becomes 100. The higher is the amount of light interacting with the pixel, the more photoelectrons there are, up to pixel saturation (clipping, well overflow).

When the shot noise exceeds the read noise, we can think of the data as to be (photon) shot noise limited.

On a side note, dark subtraction (despiking) and flatfielding are two important methods of getting rid of:
- dark noise (a combination of dark shot noise, proportional to temperature, as it is caused by thermally-generated electrons; and dark fixed pattern noise, proportional to exposure time). Dark noise is minimized through black frame subtraction, in camera it is "long exposure noise reduction", when right after the main exposure to capture the scene a camera captures a dark frame with the shutter closed and the same exposure duration used for capturing the scene,
- fixed pattern noise (caused by response and optical non-uniformity),
both methods, performed out of camera, are useful in a studio and nearly a must in astrophotography.

Among several important methods to get a higher exposure: add light / wait for light, maybe worth mentioning.

The note for adding light cannot be repeated enough. It's the photographer, who is responsible for the circumstances (as long as it is either static shot or "catching" shot).
We not only capture light, we often make it. 💪

There is also photon response nonuniformity, which is fairly well controlled in modern CMOS cameras.
There is also quantizing noise, which may or may not be well controlled depending on camera and camera settings.
There is also amp glow and black point variation.

A distinction should be made between frame to frame invariant noise, which can to some extent be calibrated out, and noise which is stochastic from frame to frame.

These days, explanations of how to minimize noise should include optimum ISO settings for dual conversion gain sensors.

I don't know of any cameras that can reliably determine that there will or won't be highlight clipping.

Thank you to those who have replied so far.

The purpose of this thread, as mentioned in my op, is to explain noise and various aspects of it in layman terms since this is a Beginners Forum.

I'm sure there will be some beginners who will be interested in and able to absorb/understand the extra technical details but I'm not sure they are what many beginners or maybe even experienced photographers think about when setting up a shot.

The concept behind maximising the quality of the raw data is very simple imo and in keeping with the KISS principle you basically just need to get as much light onto the sensor as possible while the shutter is open and within dof and blur constraints without clipping important highlights.

Well, I suppose it depends on the scene lighting and the camera settings.

If I set f/2.8, 1/30s and ISO 400 on a fully sunlit beach scene I'm pretty sure most, if not all, cameras' live histograms and/or LiveView will tell me with just about total certainty there will be clipped highlights in the data.

Personally, I don't need absolute certainty from the camera. As I mentioned earlier, in my experience the camera's histogram is a very good "near enough" indicator of clipping for both raw and jpeg data.

If you really want to keep it simple, ignore read noise.