Reset transistor is switched on to set bias on the photodiode = Vr This removes the photo-charge accumulated during the previous exposure and defines the start of the subsequent exposure.
During the pixel reset phase the gate of the reset transistor is pulsed high causing the channel of the transistor to conduct. The channel behaves very much like a low-value resistor and as such generates "Johnson Noise" that originates in the random thermal motion of electrons flowing through it. This noise has a flat spectrum but is bandwidth limited by Cn the capacitance of the measurement node. In the case of the 3T pixel shown above, this capacitance comprises the photodiode and other parasitic capacitances that appear in parallel with it.
When the reset pulse goes low the transistor is switched off and the instantaneous value of the noise waveform is "frozen" onto Cn where it adds a small random DC offset to the voltage across the photodiode.
Reset noise can be the dominant noise source in an imaging system but can be removed entirely using the technique of Correlated Double Sampling (CDS). Any low-noise scientific CMOS imager must incorporate CDS except where other noise sources may dominate, for example high-flux regimes.