‚the
>. in
the
t of
can
the
has
255
na
ing
lata
S 18
dth
ent
mn
led.
ults
nts
ade
ent
1),
um
can
GB
Cruz, Santiago
FILTERING
Kernel: It 7
11-241
(b) AA 1
IMAGE BINARIZATION
(THRESHOLD ,80 DL)
— d
OF THE FILTERED IMAGE (b) AROUND
THE LINE WITH MORE PIXELS OF
| EXTRACTION OF SELECTED PART OF
255 GREY LEVELS IN (c)
(d)
Profile in a well defined edge part Profile in an edge part with noise
250 - : ; 250
200
150
100
GREY LEVEL
GREY LEVEL
50 : 50
e Qo -- Si Em
o | Sets em s
oO 5 10 15 20 25 30 0 5 10 15 20 25 30
LINE (y) LINE (y)
Figure 5. Process for the inner orientation based in edge detection.
2.3 Automatic inner orientation and coordinate transformations
Once corners have been computed, inner orientation is made following the process stated in figure 1. The coordinate
system is centered at the IPP (computed by intersection of diagonals) and the image x-axis is forced to be parallel to the
lower edge. This implies a translation and a rotation. Pixel coordinates can be transformed in mm by scaling the image
according to the pixel size. Thus any point measured in the image can be adequately referred to the IPP. But also,
because the program marks a pixel in the computed corner positions, an inner orientation by means of a conventional
digital plotter is possible measuring the marked position.
The method applied this way does not consider any film or scanner deformation. Scanner errors are suppose to be high
because the use of a desktop scanner instead a photogrammetric one, according to the low cost philosophy employed in
this work. But if exact corner positions for the camera are known those errors can be partially corrected. There are
several ways to know accurate corner coordinates. These coordinates could be considered as “calibrated” corner
coordinates. In some cameras, especially those where the magazine can be removed, the backside at the focal frame is
accessible and it can be measured by any precision instrument (a caliper or jig). Also some glass plates can be exposed
(as made by Donnelly, 1988) and then the format edges can be accurately measured with a comparator. Finally, if there
is not possibility to obtain glass plates for a small format camera, several film frames can be measured with a
comparator and then the mean values of the frame corners are considered as “calibrated” corners. In this last case, the
errors are averaged between all the measured frames. With any of this methods at least there are some reference fixed
values (“calibrated”) that can be compared with the measured/computed ones.
Coordinate transformations permit removing some film and scanner errors, With film-based images and inner
orientation by means of computed corners, Fryer (1992) proposes the use of conformal (4-parameters) transformation
instead an affine (6-parameters) one. Affine transformation can transfer and distribute errors across the entire frame
because corners are computed, not measured. But in the case of scanned images additional errors are introduced and
may be a transformation with two scale factors (affine) is closer the reality, although measurement with an indirect
method are necessary to get “calibrated” corners. The uncertainties of the correct position of corners for a particular
frame can be overcome by the high redundancy in the measurement of the edges. But anyway different errors have their
largest impact at the frame edges, such as film unflatness, deformation, distortion and others, so some cautions has to be
taken in the use of any transformation. User should decide the best transformation to employ. More information can be
found in Fryer (1992) and Robson (1992). The program presented in this work permits the inner orientation with
different transformation types. If no “calibrated” corners are present, inner orientation is reduced to center the reference
coordinate system at the IPP and to scale the image according the scanning resolution. If the “calibrated” corners have
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 153