- black/white or 256 colour (8 bits), 1200 or 2400 dpi
- true colour (24 bits), 1200 dpi.
One should note that the SVGA 15" monitor of a PC has
a resolution of 1024 x 768 pixels and a size of about 12"
x 9". The density of the displayed image corresponds to
about 85 dpi (pixel of about 300 um). Therefore, if one
represents the pixels of the digital image in screen pixels
in the ratio 1:1 (that is a pixel on the screen corresponds
to each pixel of the image, without loss of information)
one obtains a representation with the enlargements
shown in Tab 2.
The images are therefore observed without any optic aid,
with an enlargement that is usual for the operator: 7x in
the case of photographs scanned at a 600 dpi density
(aerial photographs) and 14x if the density is 1200 dpi
(ROLLEI photographs).
dpi of the Enlargement Displayed window on
original on a 15" SVGA the full screen
image screen [mm x mm]
300 3.5 x 87 x 65
600 7x 43 x 33
1200 14x 22 x 16
2400 28 x 11x18
Table 2 - Enlargement and portion of the original image
displayed in ratio 1 pixel = 1 pixel
2.2 Calibration of a DTP scanner
The wide use of the hard-copy scanning process to gen-
erate digital images from traditional photographs has led
the researchers to evaluate the geometric and radiomet-
ric performances of scanner devices. An OEEPE working
group, co-ordinated by Prof. Kólbl [2], has considered
the radiometric performances of the photogrammetric
scanners, while other researchers from different Univer-
sities are evaluating other aspects such as geometry,
repeatability, etc.
The geometric errors can be due to:
e CCD sensor misalignment (in the case of three pass
scanners), that causes shifting between the bands;
vibration along the horizontal axes;
optical distortions;
acquisition discontinuity due to the buffer capacity;
non-constant geometric resolution along the two axis.
By comparing the radiometric and geometric results of a
consistent sample of acquisitions with the calibrated grey
wedge and reseau, it is possible to define a complete
model of calibration that must be applied to each digital
image produced by the scanner.
The geometric calibration model is determined by the
following procedure:
e acquisition of a calibrated grid;
e autocorrelation by pattern recognition techniques of
the crosses (fully automatic): a reference target has
been defined (see fig. 3) and used for searching the
scanned crosses on the grid, by means of the usual
pixel and subpixel matching techniques;
50
e computation and storage of the discrepancies be-
tween the co-ordinates of the original grid and those
of the acquired image.
Target
Search area
Figure 3 - Target and search area for reseau calibration
The so acquired data are used to correct, in real time,
the co-ordinates of a collimated point by means of a
bilinear interpolation inside each square mash.
3. DIGITAL RESTITUTION UNIT
The digital restitution unit consists of a high performance
PC and screen such as that described in 2., equipped
with a system for the 3-D collimation of stereo-pairs and
of a dedicated software for orientation and restitution.
The software is performed in order to use human stereo-
scopic collimation only for an approximate solution at the
pixel resolution, this being useful for the automatic re-
finement of the collimation (see 3.1) or in order to solve
some particular cases where the digital algorithms do
not find a consistent solution (see 3.2).
The stereoscopic view is automatically controlled; there-
fore collimations, in the relative orientation phase, give
the measurements of both the x and y parallaxes and, in
the absolute orientation and restitution phase, collima-
tions only measure the x parallax.
3.1 Computer assisted collimation
During the orientation and restitution phases, all the
required points are collimated by the operator, who lo-
cates entire homologous pixels with a possible error of
1+2 pixels.
When the registration command is sent by the operator,
the software checks the correct collimation at pixel reso-
lution and then computes a sub-pixel correlation. A tar-
get window of 9 x 9 pixels from left image and a search
window of 9 x 15 pixels from right image, centred on the
homologous pixels previously defined, are used in this
phase.
If a pseudo-normal scheme is adopted for taking images,
the extraction of search and target windows can avoid
the epipolar geometry. In fact, if o, ¢, xk are < 5°, the
relative tilt between the epipolar lines can be neglected.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996
If a is a ge
the searc
shift in the
written as:
If the imac
can be wr
image is «
have to bt
unknown
computed
In numero
this sub-pi
that is 0.1
a bi-dimer
mined w
analytical
resolution
3.2 DEM :
A very us
possibility
When usit
using the
A number
given AZ,
centred or
the previo
then proje
homologo
relation cc
puted. TI
measure
value of tl
The shap
25 x 25 |
are given
25
If the res:
1 mm? c
