AMETRY
back to
side of
36, were
es, but
, Real,
results,
farconi,
d of an
onically
.), size
) a pair
signals
nix 634
with an
.erminal
ts show
optical
isidered
stem to
.ransfer
in-house
>. Real,
th some
digital
pared.
imits is
eds the
‘or
nel
processed image size reduced. In this example a windowed area approximately
one-quarter of the entire 380x480 pixel image is processed with the convolu-
tion kernel indicated. This technique is especially germane if commercially
available processors are unable to match tomorrow's 1000x1000, 50MHz solid
state camera outputs. If the unprocessed surrounding border is
objectionable in the stereo view, it may be reduced in intensity or blanked.
The alternative, frame rate reduction, may be an acceptable alternative in
some stereo measurement situations. The following crude experiment was
conducted to determine a relationship between updated video frame and object
(aerial photograph) movement.
Table 2. Object vs Frame Grab Rate
Object Frame Grab
Rate mm/s Rate (frames/sec)
2 30(V), 25 (X)
1 25(W), 20(X)
0.5 16(7), 12(X)
0.2 12(V), 8(2), 4(X)
Y = objectively acceptable X = objectively unacceptable
Experiments with a stereo field must be conducted before more decisive
conclusions can be reached. Fig. 3b shows two simultaneous image operations
displayed together, (the processing associated with each field may also be
interchanged). Window size and position is adjustable.
Parallel processing is illustrated in Figs. 4 and 5. Histograms are
shown for selected portions of an image, Fig. 4a. (In these results the
histogram result is biased to show only the significant peaks.) The histo-
gram in Fig. 4b is for the entire image, rather than the windowed portions
as in Fig. 4a. The processing kernel shown has concentrated much of the
photometric data into a narrow peak. Yet a different processing of Fig. ^a
appears in Fig. 4c with an even more pronounced concentration of illumina-
tion. It is important to note that the histogram function can be used to
measure, then modify the image based upon that measurement. Since computa-
tion of image equalization or specification, Hall, 1979, is usually
performed in a host computer, the author proposes using the measured distri-
bution peaks to determine which one of n previously stored look-up tables to
address to approximate the desired transformation in a real-time environment
in order to circumvent delay caused by including the host in the loop. Two
dimensional autocorrelation and histogramming are illustrated in Fig. 5.
The double cusps in the left hand windows represent autocorrelation of the
windowed regions along the major x, y axes of the window, Real, 1984b. The
histograms in the right hand windows are for the entire image and are biased
to show only significant peaks. Fig. 5a is the "original" with a rather
broad distribution of illumination. Fig. 5b is processed with the kernel
indicated for shading compression and edge enhancement. The autocorrela-
tions and illumination distribution are thereby compressed. Fig. 5c is the
result of image subtraction with a diagonal displacement of two pixels.
Fig. 5d is processed with the kernel shown for edge detection, which proves
to be more effective than edge detection by image displacement, Fig. 5c.
Note the narrowed autocorrelations and histogram. In these experiments, the
image processing, correlation and histogramming are parallel operations.
The windowed results are normally viewed integral with the displayed image
and are clearly visible in contrasting color in a way they cannot be seen in
a black and white reproduction.
95