In reply to Martin Hore:
> Seriously, please correct me if I'm wrong, but I didn't think image noise resulted from hot pixels.
You are correct.
There are broadly 3 sources of "dynamic" noise - that is noise that varies frame to frame in a consumer digital camera
(1) Dark current - this is a result of the temperature of the pixel array, and is a signal that builds up over time. It builds up at a different rate in each pixel, leading to a noisy look. However, it only becomes an issue on very long exposures (think astrophotography, not night time cityscapes). It can be compensated for by subtracting a 2nd image taken immediately after the first, but with the shutter closed. This is not a problem for consumer compact cameras at all. It arrises of ~ 5 second or longer exposures on dSLRs.
(2) Shot noise - quantum noise in the number of photons of light arriving per pixel per exposure. This gets worse, per pixel, as you put more pixels into a fixed sensor size because you have fewer photons per pixel per exposure, and the signal to noise of shot noise gets worse for smaller signals. However, averaging multiple nearby pixels to produce a lower resolution improves the signal to noise ratio in the averaged pixels, equivalent to having had fewer pixels in the first place
(3) Read noise - quantum noise in the electronics that convert the signal from each pixel in the sensor array into digital numbers. There is a fixed amount of this per pixel, so more pixels in an image means more of this kind of noise. It is not fully compensated for by averaging pixels.
Depending on conditions some combination of (2) and (3) dominate for modern consumer cameras (mobile phones, compacts and dSLRs)
"Hot pixels" are something else entirely - individual pixels that are borked and always report a high intensity of light. These are typically identified by looking for bright pixels in an image of very short, or zero, exposure, and then the value reported by that pixel is discarded and replaced with an average of its neighbours. This happens behind the scenes in many cameras; when you think about it if you produce an array of 20,000,000 pixels, it is very unlikely that all off them will be produced error free if there is even the smallest possibility of an error occurring in a pixel.