3.3 Read noise
Another source of noise that is inherent in all CCDs is read
noise which is considered as an important CCD noise
particularly for low-intensity images. In principle, this type of
noise can be classified as time noise, because the random
motion of a particle change over time (Howell, 2006). Read
noise is typically defined for a number of electrons in the
process of converting input signal to output voltage (Marquez,
2012) and consists of two inseparable components. First, is in
the Analog to Digital Signal conversion process, where it is
believed that every amplifier chip and A/D circuit converter
produces a statistical distribution of possible responses with the
focus on the mean value (Gaussian distribution), although the
statistical distribution of this value is not necessarily Gaussian
Thus, even in a hypothetical case, if a pixel is reading out the
same pixel twice, each time with identical charge, the answers
can be slightly different. Second, the output electronic circuits
produce fake electrons that will create unwanted random
fluctuations in the output. These two effects are combined and
create uncertainty in the final value of the pixel output. The
average value of the uncertainty is called read noise that is
controlled by the electronic properties of the amplifier range
output and output electronic (Howell, 2006).
Before computing the read noise some information about the
parameters with significant impact on the noise must be
available. Three factors affect the results, i.e. CCD gain, flat-
field image and bias frames. Zero image or bias frame enables
us to measure zero noise level from one CCD.
To avoid negative values in the output image, electronic of
CCD with one offset is set up. This amount of offset level
names bias. One common way to assess the level of bias is to
use bias frames. Method of obtaining bias frames is similar to
that of dark frames. Thus taking image should be done in dark
conditions (shutter closed) and in the shortest possible time
(depending on the camera ability). Of course if more than one
bias frame is needed, images should be obtained at the same
temperature (Berry, 2005). (Parimucha, 2005). Flat-field frames
are described in the non-uniform noise pixels.
Now let's see how bias and flat field frames can determine the
gain rate and read noise. First, a mean value of the pixels of a
flat field and bias is calculated and are respectively named F^
and B^. In the next step, the standard deviation of the measured
images (shown as 95 ando) will be calculated. The standard
deviation can be obtained from the following equation:
o= ZEN, (x — Ww? (5)
Where — N-number pixels
u =mean pixels
x;=noise in each pixel
After we obtain standard deviation of bias image and flat-field,
the following equation is used to calculate the gain:
Gain = == (6)
0g—08
Where og =variance of flat field
of =variance of bias
Finally, the read noise is calculated from the following
equation:
Read noise = eem (7)
At the end, it should be noted that adding items to amplifier
design, pixel output synchronizers and different semiconductor
designs can be used to increase the electronic output efficiency.
Different methods of producing integrated circuits help to
improve the read noise function.
4. DISCUSSION AND CONCLUSIONS
The main objective of this paper is to differentiate and measure
IKONOS sensor electronic noises from atmospheric corrected
images. Though, measuring all satellite electronic noises is hard
or impossible, in this study, it is tried to assess some important
CCD noises on their circuits that may have far greater impacts
than any other noise on the images from satellites. As can be
seen in Table 2, the DN values of dark object and noises
investigated in 4 bands of IKONOS MS images are shown as a
percentage of the amount of dark object. Noise removal
algorithm is shown in Figure 3.
Table 2. The percent of electronic noise from dark objective
pixel at four MS IKONOS bands
Dark
[Iu
DN
Dark -18x107
current
%
Non- 34% 10” 27x10 -6x10? 1x107
uniform
%
Read ESS 3x10 $x10^ 11x10
out
%
Dark Read joi E
current out fielding
noise noise Pro
Dark
object
without
Hoise
Dark
Glow dade
value
Fig. 3. Removal algorithms for dark current noise, read noise and flat-field from dark object pixel
120
Althou;
atmosp
electro
tochan;
estimat
Baltsax
Geome
for 37
Resolu
Berry,
Proces
Chrien
and Fc
Respoi
Delane
Applic
Dial, ‘
produc
Gebre:
based
as pre
KwaZ
259-21
Howe]
Editio
Obser
Kodal
Kodal
Léves
frame
R&D
Tiu, €
2008.
Single
Mach
Mahit
Comr
Engir
Marq
paran
adapt
1208(
Mula
Geok
GeoE
confe
Pagn
Radic
imag
