Image Processing Reference
In-Depth Information
3
Major Typ
es of Noise in Image Sensors
The most important aspect of the performance of image sensors is sensitivity as character-
ized by signal-to-noise ratio (SNR). From this perspective, it is important to understand
the different types of noise in image sensors. There are various types of image sensors.
The main noise type and its level depend on the sensor type, each of which has its own
advantages and disadvantages.
Noise disturbs the accurate reception of a signal value by overlapping with it. Various
types of noise occur in the time domain (one dimensional), space domain (two dimensional),
or both. Noise that fluctuates in the time domain or the time and space domain is called
temporal noise, while noise that arises at the same position in images is called fixed-pattern
noise (FPN). Since FPN is easy to detect visually, it should be suppressed with high accuracy.
On further inspection, there are many examples of noise generated in specific devices or cir-
cuitries. Since signals of general image sensors are output sequentially, temporal noise that
arises at a common circuit node overlaps with the signal of each pixel and is distributed to
the whole image. Although temporal noise appears to fluctuate in the time domain in mov-
ing pictures, it becomes FPN in still pictures because there is no time domain.
Noise classification is shown in Figure 3.1. Among the random noise that is a critical
issue for image sensors are noise caused by circuitry and transistor devices as well as
optical shot noise, which fluctuates in both time and space domains. In recent years, along
with shrinkage of the metal-oxide semiconductor (MOS) transistor size, random telegraph
noise (RTN), which is generated by particular transistors, has become a serious issue. The
level of RTN varies with time at particular pixels and has the property of both temporal
noise in the time domain and FPN in the space domain. Therefore, in Figure 3.1, RTN
is listed independently from other transistor noise, while it should be categorized as a
transistor noise. Since image sensors are basically driven by periodical clock pulses, their
leakage causes synchronous noise contamination, which appears at the same position in
images and looks like FPN. FPN is also caused by a characteristic variation between the
pixels of the image sensors themselves.
Among the types of noise, optical shot noise increases with incident light intensity, as will
be described in Section 3.4, and rules the SNR at high light illumination. Other types of noise
rule the SNR at low light levels, because they are constant and independent of light intensity.
3.1 Amplitude of Noise
Noise is fluctuation superposed with a signal. Noise varies around a true signal value, as
shown in Figure 3.2. Expressing a time-varying signal by noise and a true signal value as
s
(
t
) and
s
0
, respectively, their relation can be shown by
()
=
0
(3.1)
st s
35
