ribed
an be
nt of
with
ssed.
enen
er bis
auch
ieden
licher
ional
d to
n the
ation
inate
id y-
ative
ror is
ad by
o the
tech-
/ one
The
ional
ogies
rinci-
) one
hoto-
error
| and
ar ar-
tions
vaths
juare
ze of
t. In
these scanning approaches geometric errors can be produced
due to inaccurate assembly of neighboring swaths or tiles.
Since it is usually easier to keep the mechanical position-
ing error of the sensor low for small scanning regions than
for large areas. However different scanning technologies can
deliver different types and different quantities of geometric
errors.
4 TARGET DESIGN
Geometric scanner properties from DTP scanner to the high
performance film scanner with geometric resolutions from 400
dpi to 5000 dpi have been investigated using both, glass and
film based test targets. As a result of investigations a suit-
able high contrast film based geometric accuracy test target
is proposed, which allows automated but also interactive geo-
metric accuracy evaluation. The aim is to obtain both, global
as well as local geometric accuracy out of one digitized tar-
get. This leads to the need of an adequate spacing between
measuring marks. The higher the density of measuring marks
in the target, the better is the ability to detect errors which
can be missed otherwise. The dimension of the scanner's
built-in CCD arrays leads itself to the limitation of the width
of one scanning swath or tile. Currently linear CCD arrays
from 300 pixels up to 8000 pixels and square array CCD's up
to 2000 x 2000 pixels are used as scanner photo-detectors.
Depending on the magnification of the optics different areas
on the scanning plane are covered by the photo-detectors.
When using a target with a low measuring mark density, pos-
sible errors in between two marks cannot be detected and
therefore they can be overlooked. An extended number of
tests was carried out to determine the feasible number of
measuring marks distributed over the test target to produce
a sensible and precise geometric accuracy result for various
types of scanners. Preferably the target would be available on
a glass plate to avoid sensitivity to changes in temperature.
However, many scanners are unable to cope with glass plates.
Therefore a film based target will need to be available. For
film targets it is proposed that a five degree change in temper-
ature can cause a change in length of 12 um [Wolber, 1991].
Normally a slight change of distances between marks due to
temperature variations does not have an influence on measur-
ing accuracy obtained after a Helmert or affine transforma-
tion, as long as a global change is affecting the whole target
equally. Additionally a general useable target has to be de-
signed to support transparent as well as opaque scanning.
The marks ought to be symmetric (circles, squares, cross-
ON DD
Figure 1: A collection of possible measuring marks digi-
tized on a DTP scanner. From left to right: circular mark;
quadratic mark; cross-hair; circled cross-hair; cross-hair with
arrows.
hair, see Figure 1) and their coordinates need to be measured
on a precision instrument, e.g. an analytical photogrammet-
ric plotter. Generally, the shape of the measuring mark de-
pends on the type of evaluation process. The circular and
square marks are better suited when evaluating by means of
a center finding algorithm. Matchig of point positions by
183
correlation is a technique where symmetric marks, especially
filled circles or squares are to be preferred. The more com-
mon cross-hair is perhaps desired for manual assessments on
a computer screen. Former investigations show that measur-
ing marks to be centered by a centering algorithm must have
a diameter of at least 4 pixels to achieve a centering accuracy
which is sufficient [Trinder, 1989] [Maalen, 1993]. This has
to be taken into account also when investigating low resolu-
tion scanners, where it leads to the need of relatively large
measuring marks. It is of main importance for evaluation by
an algorithm to ensure high contrast which allows that the
measuring marks are distinct from a target's background and
therefore can be segmented easily. As a result we have de-
signed a target as shown in Figure 2, suitable for the entire
range of low resolution DTP to high resolution film scanners,
flat-bed and drum devices, opaque and transparent systems.
Figure 2: A subsection of the high contrast geometric ac-
curacy target containing 1440 (48x30) measuring marks uni-
formly distributed over an area of 23cm x 16cm. The distance
between neighboring marks is 0.5cm. The dimensions of the
quadratic marks are 0.34mm x 0.34mm.
5 VISUAL ANALYSIS
The center of a digitized mark has to be found visually by
placing a cross-hair cursor at the desired location of the mark.
This requires the availability of a zooming function to find the
mark with sub-pixel accuracy. It should be possible to de-
termine the center manually with an RMS error of typically
5 % of a pixel [Trinder, 1987]. Manual cross-hair pointing
precision can deteriorate as target sizes increase. Additional
errors can occur due to asymmetric distribution of the pixels
on the target and due to low signal to noise ratio (SNR).
The geometric accuracy of the scanner is then obtained from
computations comparing the measured positions of digitized
marks with those known on the analog original. However,
due to the significant effort for manually measuring a large
number of marks (e.g. 600 marks) on a computer screen it
is essential that visual analysis only should support the auto-
mated investigation. Support can be needed for verification
of unusual, dubious or obviously false results obtained by au-
tomatic evaluation.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996
