random displacement of the image elements due to
micro relief within the scan pattern, and (2) photo-
graphic grain. Also the video signal-generation process
at the PMT’s introduces quantum noise because of
the high bandwidth of the process and the limited
illumination which can be provided with reasonable
CRT life.* Under conditions of low contrast and
high-background density in the imagery, PMT noise
often becomes the predominant output of the PMT's;
that is, the video signal-to-noise ratio may be substan-
tially less than unity.
As was noted, each video-correlator channel is
basically a cross-correlator. These cross-correlators
produce control signals which are relatively noise-free
by processing noisy video signals. Stated simply, a
cross-correlator performs the mathematical operation
T
Cr = fs ovo oa (1)
where C(7) is defined as the cross-correlation between
the video signals, v,(t) and vy(t), and 7 is the time
displacement produced by a position displacement
Ax or Ay. Basically, a signal-to-noise ratio improve-
ment is obtained in the correlator channel because
the noise present in the individual cycles of v, and v,
is averaged over a time, T, which is much greater than
their average period.
As Equation (1) indicates, C(7) is a function of the
amplitudes of the video signals v, and v, as well as
the time difference 7. To eliminate the dependence
on signal amplitude, the video signals in the auto-
mated analytical stereoplotter are normalized using
automatic gain-control (AGC) amplifiers. The AGC
amplifier outputs are passed through bandpass filters
and converted to binary wave trains so that the re-
quired multiplications can be performed in simple
logic circuits.
Because of the above equipment features, equation
(1) gives only general information about the actual
video correlator outputs. To obtain more complete
information, it is also necessary to consider the in-
formation content of the imagery, the spatial and
temporal bandwidth limitations of the flying-spot
scanners, and the overall signal-to-noise ratio in each
video channel. A complete mathematical analysis of
the cross-correlation channel gives:
*An accompanying paper (J. J. Edmond, “Orthophoto Genera-
tion") gives a more complete discussion of flying-spot scanner noise.
Significant noise sources not mentioned here are CRT wear patterns
and phosphor noise; however, these can generally be removed by a
"leveler" system which employs a separate PMT to detect CRT output
fluctuations and electronic circuitry to compensate the video signals.
C(Ax) =
©
DU. ] ; WAX
Loin! [in 2 d
= sin Bh foe Gyy(w) cos v 2 ;
Ay=0 (2)
where
C(Ax) = average cross-correlation for Ax relative
image displacement
H; (jw) = transfer function of the electro-optical
system
Gyy(w) = cross-power density spectrum for the
photographic imagery
e; -[S? to, 2]. i7 122
S? - mean-square signal
c,;? 7 mean-square noise in video channel i
v 7 scanning-spot velocity
For the condition Ax = 0, implying that the images
are matched, the integral in equation (2) becomes
equal to the mean-square signal, S2, and the term
within the brackets becomes S2/o,0,, à simple func-
tion of the video signal-to-noise ratio. C(Ax) is thus a
measure of the image quality, as seen by the corre-
lator at a particular point in the stereomodel.
A similar analysis for a parallax channel gives:
P.(Ax)S
2 1 2 AX
C2
umi 53/2
x sin i; fien G, y (c) sin 7 tal.
—00
Ay =0 (3)
where P, (Ax) is the average x parallax for Ax relative
image displacement. In examining equations (2) and
(3), note that the major difference is that C(Ax) con-
tains a cosine term while P, (Ax) contains a sine term.
This difference results from the quadrature networks,
and produces an even function for C(Ax) and an odd
function for P, (Ax), as shown in Figure 3.
Most automatic stereoperception systems contain
special features which adapt the scanning and correla-
tion process to the particular requirements of photo-
grammetric instruments. An example is the computer
control of the scan generator in the automated ana-
lytical stereoplotters, noted in the discussion of
Figure 1, in which the size and shape of the scan
patterns for each photograph are controlled. The
scan-shaping process implements a first-order trans-
formation of a reference model-coordinate scan pat-
tern to photo-coordinate patterns for each photo-
graph. As indicated in Figure 4, this transformation
compensates the scan patterns for perspective distor-
CHAPELLE, WHITESIDE, AND BYBEE
Sho b oa 4n AN be PA UA