Figure 6 Top row: left (left) and right (middle) SPOT image at 4th pyramid level and normalized 
difference image (right). Bottom row: binarized difference image (left), image with selected 
points (middle), image with cleaned selected points (right) 
were projected onto the search image by using an average 
height of 1000 m, and these pixel coordinates were 
compared to the known ones. The RMS differences were 
32 pixels in x and 2 pixels in y, with the maximum error 
being 72 and 5 pixels respectively. Thus, a refinement of 
these approximations by an image pyramid approach is 
necessary. Àn alternative approach would be to actually 
transform and resample the search image by using the 
PMFs and the average height. In this case, the 
disadvantages are (i) the computational costs for the 
transformation and the resampling, and (ii) the 
degradation of the data. The advantages include: (i) 
matching can be performed using only shifts, thus 
resulting in computational gains which in case of many 
points exceed the loses, (ii) detection of radiometric 
differences can be applied as proposed above, and (iii) 
since the y-parallax of the co-registered images is very 
small, the images can be viewed stereoscopically (which 
anyway is required in digital photogrammetric 
workstations). 
6. ACCURACY TESTS 
The accuracy of the matching algorithm was tested by 
using the 25 m DTM of Switzerland which is generated 
by the Bundesamt für Landestopographie. The DTMs of 
the 1:25,000 map sheets 1224 and 1225 were acquired. 
Each DTM has 701 x 481 nodes in E-W and N-S 
direction respectively. The DTMs are produced by 
bicubic interpolation in x and y direction, whereby the 
known heights are supplied by digitised contours, lake 
contours and spot heights. The accuracy of these DTMs 
was checked by bicubic interpolation of the heights of ca. 
1000 spot heights and comparison to their known values 
which have an accuracy of 1 - 2 m. The 1224 DTM was 
derived from ca. 107,000 height values and has an 
accuracy (RMS) of 1.9 m. The height range is 900 m but 
the terrain is generally smoothly changing. The 1225 
DTM was derived from 252,000 height values, has an 
accuracy of 4.1 m and a height range of 1500 m. 
Although it is not the most extreme case that can be 
encountered in Switzerland, the terrain is in most parts 
steep. Forests cover ca. 2096 of map sheet 1224 and 35 - 
40% of map sheet 1225. In the latter there are also lakes 
covering ca. 4% of the area. Some clouds were present. 
The radiometric differences were larger in map sheet 
1224 which included agricultural areas. 
The aim of this test was to check the accuracy potential of 
the algorithm. Thus, good approximations derived from 
the given DTMs were used. The measurement points and 
their approximations were derived as follows. First, an 
orthophoto for each DTM using one of the SPOT 
gradient magnitude images was generated. The points 
were selected in the first pyramid level of the orthophoto 
by using a thin-out window of 3 pixels (27183 and 26064 
points in the map sheets 1224 and 1225 respectively) and 
were projected into the original orthophoto images. After 
exclusion of the points at the areas of radiometric 
differences through subtraction with the orthophotos 
from the second SPOT image, 20,180 and 22,592 points 
remained. The X, Y coordinates of these points were 
readily available (since the images were orthophotos) and 
the height was bilinearly interpolated from the given 
920 
  
DT 
coor 
Forr 
-— 4 ^ Dd a. 
All: 
exer 
cho: 
inve 
wor 
con: 
mag 
and 
tran 
case 
and 
dete 
they