Why are my photos noisy - discussion

Another way to minimize visible noise is to use frame averaging, either in-camera with a tripod or handheld with alignment and merge in the post.

Thank you for your thoughts Jim. Whenever I describe noise and its aspects in a beginners environment I always prefer to mention read noise but I make it clear it is usually only a small component of the total noise, as I have in my op.

I haven't yet come across a beginner who hasn't been able to understand that the internal workings of a camera can introduce noise into the data.

Yes that is true but imo it is always best to initially record/capture the minimum visible noise in the data for each frame which is the purpose of this thread. You can then reduce visible noise even further using the technique you refer to.

Yes, frame averaging is a technique best applied after maximizing exposure. It has a similar effect as increasing the amount of light hitting the sensor and is different from applying NR in the post.
It can be used to overcome the limitation of the sensor size or the constraint of shutter speed.

Looking at the histogram is pretty reliable. If there is clipping, the RHS ends in a peak. If there isn't clipping, the RHS dies out without a peak.

The clipping in the histogram does not correspond to clipping in raw data unless the camera supports raw histograms (most cameras do not).

Yes that is true but if the camera's histogram shows that some of the data is very close to or at clipping point then that means the raw data is very close to clipping as well and for me that is a more than good enough indicator of raw data clipping.

Normally, I can get only about 1/2 a stop of extra exposure* before the raw data begins clipping and in just about every case, for me, that extra 1/2 stop makes no visible difference to the visible noise. So I am comfortable to use the histogram as a very good "near enough" indicator of clipping.

* exposure - amount of light striking the sensor per unit area while the shutter is open

There is no doubt that a histogram is useful.
But the discussion is whether histogram can reliably determine whether there will or won't be highlight clipping. It cannot.

There can be clipping in raw when the histogram shows there is none, and there may be no clipping when the histogram shows there is clipping. In my book, the histogram cannot reliably determine that there will or won't be highlight clipping.

Using a histogram (without UniWB) is a bit of black magic: a combination of histogram data, JPEG settings, and your experience of how it reacts in certain scenes. Also, wearing a rabbit foot sometimes helps 😆.

Having the in-camera luminance histogram showing that JPEG luminance is close to clipping does not mean that the raw data is close to clipping. The reflectance spectrum of the subject matters here. The spectrum of the illuminant matters here. The camera model matters here.

Do you regularly look at the raw histogram to see how well you're doing?

True. The Q2M does. The Betterlight backs do. Canons with Magic Lantern loaded do. None of those are likely to be the choice of a beginner.

With low light photography, "Shot Noise" (the noise inherent in the quantum nature of light) is often the major factor in image noise. Increasing the exposure is often the best way to reduce the effect of Shot Noise. (technically, it's increasing total light captured, but if you keep the sensor size constant, increasing exposure also increases total light captured).

Frame averaging is a technique for increasing exposure.

Consider a situation where you take sixteen captures of the same scene and average them. You have literally captured sixteen times (four stops) more light.

When you take four captures of the same scene, the "signal" should be the same four all four captures. Thus, averaging doesn't affect the signal. However, the noise is random. In some captures the noise will add a positive offset to the signal, and in some shots a negative offset. When you take lots of captures and average them, the noise averages out.

Averaging sixteen shots should give you the same Shot Noise as taking one shot with four stops more light.

Yup.

blog.kasson.com/the-last-word/an-mf-camera-in-your-jacket-pocket/

Yes, and thanks for writing a detailed post. It is a technique for increasing exposure.
My dream would be a camera that does frame averaging and image alignment and produces a single image. That would allow me to take images with shutter speeds as long as I do not get cramps in my arms.

Phase One makes such a camera. Not sure about the image alignment part, though. Bring money.

No alignment with Phase One. Olympus m43 cameras have frame averaging (Live ND), and multiple exposures of Nikon's DSLRs also work as frame averaging. Still, no camera can align images, meaning tripod use only, which means base ISO where the noise is not much of an issue (except with m43).

Reason #1 - There is not enough signal!

Data information capacity is a function of the system resolution and the data signal-to-noise ratio. Besides adding more light to the scene, the only way to increase signal level is to use a longer shutter time and, or wider lens aperture. The photographer is able to minimize the perceived noise level by optimizing exposure. A definition for optimum exposure could be the shutter time and, or lens aperture were set such that the brightest regions in a scene required to produce an aesthetically acceptable rendered image are exposed just below the point where the photodiode response becomes non-linear or exceeds the full-well capacity. This implies specular highlights in bright sun and bright point source lights in night scenes are intentionally over exposed. It assumes that camera ISO setting is determined solely by the longest practical shutter time and widest practical lens aperture as the ADC will become clipped if camera ISO setting happens to be to high.

Reason # 2 - Quantum mechanics.

Converting electromagnetic radiation to electrons (or photoelectrons if you prefer) is inherently noisy. In this situation the noise can be described as an unavoidable degree of uncertainty in the number of electrons produced when the shutter was open. The uncertainty is unavoidable because converting electromagnetic radiation into electrons mean coherent quantum states (the light) are affected by the incoherent nature of the detector (the photodiode). The word coherence describes a quantum state that is perfectly ordered. When light interacts with the photodiode, the photodiode thermal energy (disorder) destroys the coherence. This non-deterministic event is described by a range of probabilities which means the number photoelectrons fluctuates for different events even though the light energy is constant.

Good answer, but might go over some beginners' heads.

Jim,
Are there situations where the camera histogram does not show clipping, but the raw data is clipped?
(For example, my camera only has a luminance histogram, but I'm photographing a male cardinal)
Thank you
Sherm