rea of
'e the
lation
is +
nd 87
ction:
when
mea-
leigh-
ption
rans-
rector
n.
<x
3
Nr.
RESEAU
2
~
3
ICE
- * A 4 "e - à 4 - + + 4 - ‘
Ser vxrnd oul tit, BMS= 0.26 pixel / — X02 um, Heald. tact=140
max. devlation horizontal = 4.19 pixel / 35.82 pon
mx, deviation = vetionl = CAS pixel / 0 7.55 gen, Pixeisiiex 8.5 ques
Figure 5: Geometric accuracy investigation of a square ar-
ray CCD scanner. A subsection of the resulting residual er-
ror vectors shows the accuracy of the alignment of adjacent
tiles. If tiles are put together inexactly it can be seen from
direction and dimension of error vectors. Affine transform;
Pixelsize—8.51m; RMS error-zx 3.1um; error vector magni-
fication: 140.
7.4 Number of measuring marks
Table 1 shows the dependence of error vectors from the num-
ber of measuring marks lying in the investigated area. It can
be seen that at least a certain number of marks is necessary to
make a meaningful evaluation of geometric scanner accuracy
by use of linear conformal or 6 parameter affine traansforma-
tion. In this case of global geometric accuracy evaluation the
investigated area covers more than one of the tiles made by
the scanner for digitization.
Scanner trans- res. RMS max. max.
form. err. hor. vert.
um um um Um
DTP helm. 63.6 108.0 207.1 161.9
Scanner 1 aff. 63.6 16.4 30.5 45.2
DTP helm. 42.3 96.9 156.1 235.4
Scanner 2 aff. 42.3 35.0 74.5 87.7
Drum helm. 10.4 17.8 29.8 25.4
Scanner aff. 10.4 6.3 15.3 11.8
Photogr. helm. 7.5 7.3 9.8 19.42
Scanner 1 aff. 7.5 6.2 15.2 14.6
Photogr. helm. 8.5 4.75 9.2 9.7
Scanner 2 aff. 8.5 3.2 9.0 7.43
numb. trans- RMS max max
of form. error horiz. vert.
marks
pixel / um pixel / um pixel/ um
19x30 helm. 0.5/4.7 1.1/9.1 1.1/9.1
aff. 0.4/4.1 1.1/9.1 0.9/7.4
10x15 helm. 0.5/4.4 0.7/5.9 1.4/12.1
aff. 0.4/4.1 0.7/5.7 1.3/10.7
5x8 helm. 0.5/4.4 0.6/5.0 1.1/9.6
aff. 0.4/4.1 0.5/4.3 1.2/10.4
3x4 helm. 0.4/4.1 0.6/5.2 0.9/7.5
aff. 0.4/4.0 0.3/2.9 0.7/6.0
3x3 helm. 0.3/3.5 0.4/3.5 0.5/4.7
aff. 0.3/2.3 0:3/2.3 0.4/3.6
Table 1: Residual error vectors according to different numbers
of test marks distributed evenly. Affine and Helmert trans-
formations are used for evaluation. The investigated area is
about 9 cm x 15 cm at 3000 dpi geometric scanner resolution.
It is evident that the RMS errror decreases as the redundancy
in the number of measuring marks decreases.
185
Table 2: Residual error vectors in um according to different
scanners. Evaluated were two low cost DTP scanners and
three high performance film scanners. The target measuring
area utilized for the testscans is 11 cm x 16 cm except on the
Drum Scanner where a reduced measuring region of 5cm x 5
cm is used. DTP Scanner 1, DTP Scanner 2 ... single line
CCD principle; Photogrammeric Scanner 1 ... Multiple swath
line; Photogr. Scanner 2 ... Square Array CCD.
Scanner trans- res. RMS max. max.
form. err. hor. vert.
pixel pixel pixel pixel
DTP helm. 63.6 1.7 3.3 2.5
Scanner 1 aff. 63.6 0.3 0.7 0.5
DTP helm. 42.3 2.3 3.7 5.6
Scanner 2 aff. 42.3 0.8 1.8 2.0
Drum helm. 10.4 1.7 2.9 2.4
Scanner aff. 10.4 0.6 1.5 13
Photogr. helm. 3:5 1.0 1.3 2.6
Scanner 1 aff. 7.5 0.8 1.9 2.0
Photogr. helm. 8.5 0.5 1.1 1.1
Scanner 2 aff. 8.5 0.4 14 0.9
Table 3: Residual error vectors in pixel according to different
scanners. DTP Scanner 1, DTP Scanner 2 ... single line
CCD principle; Photogr. Scanner 1 ... Multiple swath line;
Photogr. Scanner 2 ... Square Array CCD.
7.5 Comparison of different scanners
Error vectors for conformal transformation and affine trans-
formation obtained when evaluating the geometric accuracy
performance of different scanners can be seen in Table 2 and
3. Additionally the tables show the maximum errors in hor-
izontal and vertical scanning direction. The resulting error
vectors show significant size variations for different types of
scanners. The geometric accuracy of high accurate, high
performance photogrammetric scanners surpasses the accu-
racy delivered by the evaluated DTP and prepress scanners.
Although the investigated drum scanner was expected to be
geometrically inaccurate due to its working principle, its resid-
ual errors after an affine transformation lie in the range of the
high performance scanners which have been evaluated. How-
ever, the evaluated area was only 5cm x 5cm and a lack of
flatness of the film target may thus not be effective. The
error may increase in a larger scanning area. When compar-
ing the RMS errrors for Helmert and affine transformations
it is evident that the high performance scanners produce er-
rors which lie in the same range, indicating the absence of
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996
