Ud mba u TU Re 
For the orthophoto production the USGS 30 m DTM was 
used. The accuracy (RMS) was within 40 ft, and for some 
scenes less than 30 ft, against the USGS 7.5 min. 
topographic map. 
Digital orthophotos were also produced with the above 
mentioned SPOT data in W. Switzerland (Figure 4). The 
25 m Swiss DTM of the 1:25,000 map sheet 1225 (see 
Figure 1) having an RMS accuracy of 4.1 m and covering 
an area of 17.5 x 12 km was used for the orthophoto 
generation. The height range was 680 m - 2150 m. 
Orthophotos were produced using both images of the 
stereo pair and four different versions. In all orthophotos 
the orthophoto pixel spacing was 8.333 m and the 
interpolation bilinear. The implementation of the PMFs 
was either by the incremental approach (fast version) or 
by using equation (4) (slow version). 
Version 1: DTM densification by factor 3, projection 
using basic PMFs (from linear model), fast 
Version 2: DTM projection with basic PMFs (linear 
model), threefold densification of anchor points, fast 
Version 3: like version 1 but slow 
Version 4: like version 3 but using extended PMFs 
from quadratic model 
A version like version 3 but with extended PMFs from 
the linear model gave the same results as version 3. A 
version like version 4 but with anchor point densification 
instead DTM densification gave the same results as 
version 4. 
The accuracy of the orthophotos were tested in two ways. 
The first test was a relative (inner) accuracy test 
comparing the orthophotos of the same version from the 
two SPOT images. Corresponding points should ideally 
have the same pixel coordinates. By using template 
matching the actual parallaxes (i.e. planimetric map 
coordinate differences) were detected. Version 1 showed 
increasing parallaxes of up to many pixels (mainly in x) 
while moving from left to the right part of the orthophoto. 
Version 2 showed a similar behaviour but less 
pronounced. Versions 3 and 4 showed a very good 
(subpixel) closure between left and right orthophoto, 
whereby version 4 was slightly superior in y. The reason 
for this behaviour lies in rounding errors that in the case 
of the incremental approach are accumulating. Since the 
processing starts from the top left corner and proceeds 
columnwise, it is clear that the errors for versions 1 and 2 
increase in the right part of the orthophoto. This increase 
is enhanced by the fact that the heights are larger in the 
right part of the image. Version 2 gives better results than 
version 1 (and it is also faster) because less DTM points 
(by a factor 9) are transformed with the PMFs and thus 
less errors are accumulated. The errors are larger in x 
than in y because the PMF coefficients in x are larger 
than in y. To reduce these errors (a) double precision 
variables should be used (which was the case in the 
current implementation; however, the heights and the X, 
Y coordinates of the DTM origin are 4-byte real 
variables), and (b) the first pixel of each column should 
be computed by using equation (4). Thus, errors can 
accumulate only within one column. By computing 
strictly (i.e. by equation (4)) the pixel coordinates of not 
the first but the middle line of the orthophoto, the 
accumulation of errors within one column can be reduced 
to half its length. If these precautions are not met and the 
optimal accuracy is sought for, then the slow version 
should be used. In this case it is irrelevant whether DTM 
or anchor point densification is used, but the latter should 
be preferred as it is faster. 
The second test checked the absolute accuracy and used 8 
out of the 136 points that were imaged in these 
orthophotos. Their known pixel coordinates were 
transferred from one of the SPOT images in the 
orthophotos through template matching. Thus, 
planimetric map coordinates were derived, the heights 
were bilinearly interpolated from the DTM, and these 
values were compared to the known values. The accuracy 
measures are listed in Table 3. 
Table 3 Accuracy measures (differences) for different options of orthophoto generation 
SPOT absolute maximum average 
"T 
image X Y X 
1 I5 
Y 
absolute average RMS 
X Y X 
3. 11. 
right 32.8 17] 52 72 
left 7.5 14.6 14 7.3 i 7.9 4.5 
right 182 15.0 -6.8 3.7 6.5 
left 9.7 14.6 1.9 77 8.0 4.8 
right 12.8 14.0 37 27 6.2 6.8 
left 10.6 14.1 3.0 72 7.6 5.5 
right 15.0 142 4.8 4.5 72 78 
  
*Orthophoto from left or right SPOT image 
363