6 AUTOMATED ANALYSIS
When locating the center of measuring marks with sub-pixel
accuracy a wide range of techniques is proposed and these
can be grouped as centroiding, correlation, edge analysis and
shape based techniques [Trinder, 1989] [West, 1990]. Auto-
mated measurement is possible with an accuracy of better
than + 0.1 pixels. With the scanner evaluation system the
center of circular digital dots and other symmetric measuring
marks can be found automatically with subpixel accuracy by
means of a weighted center finding algorithm [Trinder, 1989].
According to the former investigations with a target diameter
of 4 pixels of [Maalen, 1993] the automated algorithms de-
liver the best results. Target sizes of less than 4 pixels have to
be avoided due to large deterioration of precision, whereas the
precision is slightly lower but sufficient for more than 4 pixels
of diameter. The dynamic range of the measuring marks will
not significantly influence the accuracy of centering as long
it has more than 2 bits of radiometric resolution . Thresh-
olding can be used to isolate the background from the target
which results in a certain loss of radiometric range. Machine
centering at suitable measuring marks is preferred since in
the long term it offers higher accuracy, independent of the
operators skill and of the exact diameter of the targets. Fi-
nally, of course, it will also be faster to center automatically
by machine than to accomplish this by hand.
7 EXPERIMENTAL RESULTS
Various investigations of geometric accuracy on different
scanners from the low price DTP scanner to the high per-
formance film scanner were performed by means of the
SCANEVAL system. Here the individual manufactorers and
scanner models are kept anonymous to emphasize the test
technology, not the specific test result. Typical geometric
errors according to the scanners scanning principle can be
obtained by scanning and analyzing the geometric accuracy
target.
7.1 Single line CCD Scanner
Figure 3 shows the error vectors for a DTP scanner using
single line CCD technology. The change of vector orientation
in the scanning direction indicates that the speed of the CCD
linear array is not stable when moving over the scanning area.
Consequently the resulting real pixel size is varying over the
scanning region and is directly proportional to residual varia-
tions.
7.2 Multiple swath line CCD scanner
The residual error vectors obtained after an affine transforma-
tion for a widely distributed high performance flatbed scanner
can be seen in Figure 4. In this case adjacent scanning swaths
are stitched together to one large digital image. The changing
vector orientations in scanning direction indicate mechanical
misalignment of the line array CCD when scanning neighbor-
ing swaths.
7.3 Square array CCD scanner
The residual error vectors for a high performance film scanner
working with the square array CCD principle are shown in
Figure 5. To avoid geometric errors adjacent tiles have to be
assembled geometrically and radiometrically.
184
lle: pat gri ; .B2 pixel / 34.66 um, ^ Resid.foct- 54
mox. deviation horizontal = 1.88 pixel / 79.50 um
mox. deviation vertical = 2.04 pixel / 86.31 um, Pixelsize= 42.3 um
Figure 3: Residual error vectors when evaluating a test area of
11cm x 16 cm, including 19 x 30 measuring marks. Here the
scanning resolution is 600dpi (42um). The r.m.s. deviation
from ideal mark position obtained by affine transform is +
35 pm. Maximum deviations are 74 um horizontally and 87
pm vertically. error vector magnification: 54; scan direction:
left to right;
1
A
3d pef N fs + OR NT
sonia na a ee Lys I
: A uiu ; Uu T J y s. + i CE ais i
SH Es d AN 17 anid qo s "S Le eid ito ;
pobres ra
dm TR TR:
Ee A 5h. } 7 . | For 2 Je ] QE s SERA ] ] 4 =
| ES TP Ly CX Du
/ ku / E j yi | 4 id YN s OM:
A d zx 7 / / ja ed | ui.
J A ; j ssa iy. ide rid Cory 1 % œ
5 SAI a gc | ig 1 xd PN $e j UY ) v AL Ln T
pil Fh t leg ow ep epo és
A ada
-
File: pstgrid2— 1.tst.tiff, RMS- Q82 pixel / 616 pm, Resld factz 332
mox. deviation horizontal — 1.38 pixel / 10.18 um
mox. deviation vertical = 2.00 pixel / 15.02 um, Pixelsize= 7,5 jam
Figure 4: A subsection of resulting residual error vectors when
evaluating a test area of 11cm x 16 cm, including 19 x 30 mea-
suring marks. Orientation of error vectors changes in neigh-
bored measuring mark columns. This leads to the assumption
that adjacent scanning swaths are misaligned. Affine trans-
form; Pixelsize=7.5um; RMS error==4 6.2um; error vector
magnification: 332; swath scan direction: top to bottom.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996
The: vxe
max. de
mux, de
Figu
ray
ror \
tiles
dire
Pixe
ficat
7.4
Tab
ber
bes
mak
by L
tion
inve
the
Tat
of 1
forr
abc
It is
in t
