he whole 
selines of 
2 m; 
aseline of 
nd same 
aseline of 
minimum 
nplify the 
of 1/20 of 
ave been 
ce at the 
js where 
5 and the 
therefore 
ither for 
ight have 
ite some 
signalize 
pointwise 
it loading 
her than 
nponents 
> grid of 
ive paper 
oint. The 
ameter of 
pared on 
selecting 
ensure a 
X Vision 
lic are as 
sparency 
ss scan, 
) dpi in 
standard 
, gamma 
or freely 
possible 
tric and 
e been 
las been 
switched 
r of 15% 
ity of the 
is better 
A radiometric calibration has been performed by using a 
grey wedge chart; a certain non linearity and a loss of 
sensitivity in the dark region have been highlighted. To 
assess the influence on the target location by l.s.m., two 
tests have been carried out adjusting a block of 4 images 
with and without corrections: no significant changes to 
object coordinates have been found, therefore no 
corrections to g.v. were applied later. 
3.2 Geometry 
Since our scanner was not equipped with a transparency 
module, we could not use glass reseau plates. A simple 
reference grid was generated by cutting a reseau with a 
spacing of 1 cm on a 12x11 cm area (just larger than the 
size of our enlarged pictures) over a stable polyester film. 
The coordinates of the crosses were measured on a PK1 
Zeiss comparator and, after scanning, by l.m.s. on the 
digital image, using an artificial template. The two sets of 
coordinates to compare are relative to reference systems 
which may differ either in orientation and scale (in two 
directions): an affine transformation has therefore been 
estimated, assuming as observations the pixel 
coordinates. Since the interior orientation will be 
performed by using the camera reseau crosses, the 
computed transformation parameters will not be used 
later. Nonetheless, it is interesting to look at the residuals 
of the affine, since they show deformations which will not 
be recovered by the transformation in the interior 
orientation. 
  
  
Figure 3 - Residuals of the Y component (along scan) 
Figures 2 and 3 show the behaviour of the residuals of 
the X (across scan) and Y (along scan) components. It 
can be seen that distortions in X component are relatively 
uniform in Y direction, with a maximum value of about 3 
pixels. Residuals of the Y component on the contrary 
show a wavy trend and are less predictable, but also 
smaller (up to 1 pixel) than the X component. They may 
be due to some irregularities in the movement of the 
optical system. 
Another problem, occurring only when scanning large 
areas at high resolution, was perhaps due to the limits in 
memory (either of the scanner's buffer or of the PC's 
RAM): in such cases, the scan is completed in several 
stages, where the moving stage stops and then restart, 
after going back a few centimeters to reach the steady 
speed. In this operation losses up to ten rows have been 
noticed scanning a pattern of diagonal lines. In order to 
cope with this problem, thanks to the limited size of our 
images, we simply put them always in the calibrated 
area, where, at the selected resolution, the problem didn't 
show up. 
4. INNER ORIENTATION 
The transformation between pixel and image coordinates 
of the reseau crosses will partly adsorb the deformations 
due to scanning, to shrinkage of the negatives, to the 
enlargment and to the shrinkage of the enlarged copies. 
As outlined above, in principle we have at hand two sets 
of reference values: the results of the geometric 
calibration of the scanner and the coordinates of the 
camera reseau from the calibration certificate. The 
enlarged pictures proved to be sensitive to lack of 
orthogonality of the optical axis of the projector's 
objective with respect to the stages carrying the original 
and the enlarged picture. In order to account for this fact 
and to combine both reference sets we opted for a 8 
parameter transformation for the inner orientation, 
followed by an interpolation of the residuals with a 
polynomial function modelling the scan deformation of 
the X component (see figure 2) as follows: 
= 3 2 
u=a,x" + a,x* +a,xy+a,x+ asy+ a 
— 3 2 
V=b,x" + bx. *b,xy *b,x* bsy * be 
The performance of this two stage procedure have been 
compared against a simpler one (applying the 8 
parameter transformation only) in one of the block 
adjustment performed. Taking as performance index 
sigma naught and the rms of the residual on check 
points, it turned out that the double stage procedure was 
slightly worse than the single stage. This may perhaps be 
attributed to instability of the scanner, so that the 
calibration procedure should be repeated just before 
scanning. In the adjustment of all blocks therefore the 
simplest procedure has been used. 
5. POINT TRANSFER AND TARGET LOCALIZATION 
After digitization of the images, a combination of an 
automatic and an interactive approach was used to 
locate approximately the targets and the reseau crosses 
on the images for the subsequent refinement by l.s. 
matching. 
e Targets were approximately localized in the nadir 
image by first applying a smoothing filter followed by 
Foerstner's interest operator: with appropriate 
211 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996