Cameras: Sensor Noise
Noise in a camera has a strong influence on the maximum responsivity and the dynamic range of a camera. Reasons for sensor noise are quite different and are explained below:
The expression temporal noise summarizes all noise sources that have an influence on the temporary progress of a pixel value, i.e. from image to image. Temporal noise is made up from photon, dark current, readout and quantisation noise described below. It tends to be the dominant noise source found in CCDs as the charge of each pixel is shifted many times during readout.
A source for sensor noise is the light itself. The photon flux that strikes the sensor is not even, but poisson-distributed, causing the so called photon noise or shot noise. In addition it limits the maximum signal-tonoise ratio.
Dark current noise
Dark current noise is created by electrons that emerge through thermal processes in the pixel. The number of electrons increases with increased exposure time and temperature. It also increases with lower readout speed.
This type of noise occurs when charge (electrons) is converted into voltage (for CCD sensors also when charge is transferred through the shift registers or pixels). Typically due to more tightly coupled pixel to A/D path in CMOS sensors, this noise source is significantly lower in CMOS cameras compared to CCD.
A/D-conversion can also be a cause of noise as a voltage (continuous values) is converted into a digital value (discrete values).
The term spatial noise describes spatial variations of the intensities of different pixels when illuminated with homogeneous light. The causes of spatial noise include FPN and PRNU noise as described below. Spatial noise tends to be more dominant in CMOS sensors as the pixels are read out through different readout circuits.
Fixed pattern noise FPN (offset noise)
FPN is defined as the peak-to-peak difference between the minimum and maximum measured values for all active pixels in the array. As this should be measured with the sensor in darkness, it is also called dark signal non-uniformity (DSNU).
Pixel response non-uniformity PRNU (gain noise)
Pixel response non-uniformity is described as deviations between pixels with a fixed gain applied. This is caused by the fact that individual pixels have different sensitivity curves.
As FPN and PRNU are constant, the term "noise" is a bit misleading, as deviations can be corrected relatively easy by adding an offset (FPN) or multiplying a pixel specific gain (PRNU). Those deviations are the result of production specific tolerances or contaminant sensor substrate. Many CMOS cameras now feature correction for this noise directly built-in to the camera.