3. Digitalisation
Present technology allows computer access to the
information content of photographs. To that end
negatives must be digitized with sufficient
accuracy, to avoid errors induced by the
digitization process. For this reason we used the
high precision scanner ASTROSCAN of
Sterrewacht Leiden. The ASTROSCAN is a David
Mann Monocomparator modified to allow
computer controlled operation as a micro
densitometer. A 128 element Reticon photodiode
array is measuring the transmitted light from the
photograph. Stepping motors control the
positioning of this array in steps of 10 pm. So
pixels have a size of 10 x 10 pm and a slightly
bigger effective size of 13 x 13 pm. Real-time
application of a calibrated transformation of the
transmission measured provides for photographic
densities. From both the geometric and radiometric
point of view the ASTROSCAN is a high precision
scanner with a well documented performance.
Geometry. The repeatibility of the pixel positioning
of the ASTROSCAN is found to be 0.3 pm RMS.
This holds for the mean of a few hundred pixels.
Local systematic deviations stay below 1 pm. In
figure 2 an error vectorfield of a repeatibility test is
plotted.
Le
RS €
S A avi. ^ x » »485]
NN SUC QA e eoe ei
200mm | NN Bay sU. ^w» afe € e ee € » v v 7|
NON OR sage ew T= = 4 y »« |
N T1 u ya a * 9 Y «€ * 7 a ua |
Ne as y ue << + ee ra à» 4
ER AA 4 na a 6 CS 6 NRC 4 > à 4
150 EX f^ oa s au CR PIS « 4 S a« 7 €
[ 5 7-4 Bo. c ox 4 ^ « ^ € 4 c, 9
A € Oo. a cu T «re Y 3» » >
NR 4 0 uà c3» ao v o Se vr ^r
[NS 5 * Ay maya tt ares Cay
100 L« N ^ -3 4 uU 4 WW mr ^e 5 o m
IN * 2 7. y NECS tre. mà
[S ^k a ML e DATE FT REN
Sr A Ans WNT ete a a
IN f£ «731 A VN wye* € «SN o. a
50 ivo ea se VA S €t s]
a Am >, ww $- & 3€ * r7" >
D € r ^ T. y ^a aw Hy ete . ow»
eu LE dus rib L—À
100 200 mm
figure 2: error vectorfield of an Astroscan
repeatibility test
n
o
150
Radiometry. The measurement noise of the
photodiodes being about 0.07% of full scale the
noise in the densitometry is well below the
photographic noise, except for very black pixels,
where the transmitted light is less than about 0.596
of full scale. For such dark pixels a second effect
distorting the radiometry comes into play: scattered
223
light in the optical system. For all less
"overexposed" pixels the densitometric precision of
the ASTROSCAN has been shown to be better than
a few times 0.01%.
4. Search
For full automation of the procedure the targets
present in the images have to be recognized
automatically. The recognition results in
approximate positions of the targets.
The algorithm used to detect the targets in the
images was developed at Sterrewacht Leiden for
the detection of astronomical objects in starplates.
The first step in this procedure is the estimation of
the background. In the astronomical applications
the background consists of the pixels with the
lowest densities. Objects consist of excess density
in an area of the size of the Point Spread Function
or slightly larger. The algorithm detects sets of
adjacent pixels with a preset density difference
above the background value.
This procedure applied in close-range
photogrammetry will yield many more high density
areas (supposedly "objects") than the ones that
originate from the specially designed targets. In our
experiment just over 546 of the objects found were
part of a target image. All targets were detected.
To distinguish targets amongst all objects detected,
a set of parameters is computed for every object.
The parameters are used for a comparison with the
characteristics of a target image.
The main characteristics used are the following:
- within limits set by the size of the target images
two objects should be found at the same location
(the images of the circle and the ring of the target);
- the size of these two objects should have a fixed
ratio within bounds set emperically;
- the absolute size of the two objects should be
within limits that depend on the scale of the
photograph, which in turn is to be adapted to the
physical size of the targets photographed.
For 9% of the targets only one object of the two
was detected. Inconsistencies in excess of 35 um
for the two supposedly identical positions is found
in 17% of the cases, probably due to a large
"curvature" in the estimated background and/or
