too small. This problem can be avoided by integrating: instead
of transferring each single electron, the charge-carriers are
integrated locally over a certain period of time into a charge
packet creating energy large enough to be easily detectable as
output. The next link in the imaging chain is the transport of the
integrated charge packets from the integrating sites towards the
output of the device.
CCDs, silver-halide crystals in photographic emulsions and
even the human eye are "photon counters". While CCDs are
regular arrays of receptors of identical size, photographic emul-
sions consist of a random three-dimensional arrangement of
receptors in size significantly smaller than CCD cells. In prac-
tical applications, CCD images can be built up one-dimension-
ally (line-scanner) or two-dimensionally; all cells in the light-
sensitive part of a CCD are biased in the integrating mode. A
camera based on the full-frame CCD imager will be making use
of a mechanical (or LCD) shutter. After opening this shutter,
impinging photons will generate charges which are collected
until the shutter is closed. Nexi, the charge packets are
transferred first to a CCD output register and then io the output
stage where they can be converted to an electrical voltage. in
the case of photography integration of the effeci on a larger
number of exposed and processed crystals by means of a sgan-
ning aperture is necessary.
Research is continuing to increase the performance of salid-
state image sensors. Improving fabrication technology, optimiz-
ing design and creating new device architectures, all have the
same goal: increasing the signal-to-noise (S/N) ratio of the sen-
sor for a given application or a given resolution. The number of
signal electrons can be specified in terms of the number of
photons impinging on the sensor e the receptor area 4,4,
the quantum efficiency » and the integration time 7;, :
Fin! electrons 7 EX A receptor X nx T
The performance of imaging systems can be compared by
means of the detective quantum efficiency (DQE) which is
derived from the signal-to-noise ratios of the image input and
output processes:
DQE = (S/N) out (S/N)Yin
The number of noise electrons for CCDs can be split up into
three main noise sources :
— photon shot noise, equal to the square root of the number of
signal electrons, given by the equation above,
— noise electrons generated in the CCD channels (e.g. incom-
plete transfer, shot noise on dark current, fixed-pattern
noise, ...), all summed in ngs,
— output amplifier noise n
and written as:
out
; = : 2 2
f ieisel electrons NH eise electrons * n sens + n out)
Amplifier noise and sensor noise are independent of the input
signal; they are, however, strongly dependent on readout fre-
quency, chip temperature and integration time. Another impor-
tant noise source is the light fixed-pattern noise; this noise is
static and is related to the non-uniformities between the various
pixels of the imager, and it is proportional to the light input; a
typical value for this noise source is 1%. In addition to the
various noise sources in the imaging part of the sensor itself.
another important noise source is the output amplifier. The
small analog circuitry used to convert the electrons into a
voltage and to buffer the output voltage towards the outside
world, adds some uncertainty to the signals in the form of noise
electrons.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
Photographic emulsions are very effective in recording infor-
mation. Only four electrons are needed to form a latent image
spot on silver-halide crystals of an estimated average size of
(0.5 um)? a magnification factor of 250000000 results per crys-
tal. The DQE can be estimated as
DQE = SNR? - (signal / density variation 7 AD? / o*j
= (0.434? x 7) /(H x À x @p)
Typical values for the density variation op within an aperture A
are 0.02 to 0.03, 7 is the gradient of the straight-line section of
the applicable characteristic curve and H is the exposure in
ergs/cmn?.
DQE values for the three different types of sensors all are in the
area around 196. Optimum exposure levels vary as does the
range of exposure levels which is most limited in the case of
photographic emulsions One way of increasing the DQE is to
optimize the light sensitivity of the devices. At extremely small
light inputs, however, the detection limit of any signal is deter-
mined by the remaining noise floor. The dynamic range is also
partly defined by the noise level of the imager and the output
amplifier. Further reducing the various noise sources is just as
important as increasing the light sensitivity of the devices.
Since most noise sources have different origins, it 1s not possi-
ble to tackle them all at once; several mechanisms are technol-
ogy related, others have to do with the design of the device,
some can be minimized by the processing of the video signal,
and others cannot.
4. REFLECTANCE, TRANSMITTANCE AND DENSITY
The concepts of transmittance and reflectance are very similar
as will be shown shortly. In the case of scanning a transparent
photo, the amount of energy transmitted will be measured. In
the case of digitally recording an image, the amount of energy
reflected at the scene will be recorded. While we speak mostly
of transmittance, we mean also reflectance. Reflectance p is
defined as the ratio of flux reflected from the sample («b,) and
that reflected from the reference surface (®;):
p 0;/do,
Similarly, transmittance r is defined as the ratio of flux trans-
mitted through the sample (®,) and that incident onto the sam-
ple (®;):
7r=b/®,
Consider a piece of material of a thickness d which passes half
of the incident light: 7= 0,5. If several pieces of the same
material were placed on top of each other, on would observe
that with a doubling of the thickness half of the transmittance
would be obtained. The inverse of the transmittance is called
opacity O, and it would double with a doubled thickness as
shown in Table 1.
Table 1. Thickness of Material, transmittance and opacity
d O
|
2
4 $ 16
A human observer would sge decreases in brightness not cor-
responding to the geometric progression of opacity but propor-
tional to the thickness; the laíter js obtained by introducing the
transmission density D as the logarithm of the opacity (or of
the inverse transmittance):
D = log - log(|/r)
International
The proportic
Table 2.
Table2. TI
de
o3|t32|—
d
XIII [tA
9
10
Reflection de
Assuming th
range of 3,0,
normal 1-byt
over the valu
fore to a den
presented by
clear materia
shown in the
transmittance
13-bit and 14
Table 3 2
de
D D (8)
3.05 0
2.90 12
2.75 25
2.60 38
2.45 50
2.30 3
2:15 75
2.00 88
1.85 100
1.70 113
1.55 125
1.40 138
1.25 150
I 10 163
0.95 176
0.80 188
0.65 201
0.50 213
0.35 226
0.20 238
0.05 231
0 255
While the d
range, the t
achieve simi
13 to 14 bits
Density req
country dep:
film and th
mined by t
achieved at
illumination
