Well this all explains how noise is amplified but not where it comes from. Bifurcator that isn't correct, if it were taking 2 photos and averaging the result wouldn't halve the noise. But it does because noise is a random Gaussian distribution.
Ok so starting at the analogue digital converter and the signal amplifier. Both have noise determined by a few things such as external interference, noise on the powersupply (voltage regulators in consumer electronics regulate by switching on and off some 200000 times a second), and by Johnson noise (this is thermal noise relating to resistive elements and temperature, it exists it can not be reduced without reducing the resistance in the path).
Back a bit further we have Shot noise which dominates the amplifier. This is fluctuation in the movement of electrons through a conductor. It gets worse in a semiconductor like a MOSFET amplifier (there's one of these under each photo detector cell on your sensor).
The photo detector itself produces noise too, normally three types. Dark Current, surface fluctuation, and statistical gain. Dark current is when random thermal changes generate electron hole pairs and generate current even when no photon has hit the sensor. Surface fluctuation is a result of manufacture, photos taking a different path into the detector may not generate an electric charge than another photon. And Statistical gain occurs in avalanche devices (though I am not sure if these are used in cameras.)
All the way back there's one more source, quantum noise of the photon, which in it's most basic form is Shot noise applied to a photon instead of an electron, and in more complicated analysis is dependent on the uncertainty of the E-H field amplitude and phase of the light entering.
Now over to you Socrates so other people who didn't take 4th year photonics at uni can understand this too.