des the 
using 30 
1ethods. 
ww band 
ow band 
> but it 
cteristic 
1and, in 
ause the 
tle error 
)) 
  
  
(b)Using wide band filter (y =0.5) 
Fig.10 Results of normal filtering. 
2-4.ADAPTIVE FILTER 
Here, an adaptive filter is introduced in order to 
restrain the deterioration of the edge 
characteristic. This is a filtering method which 
automatically select to suitable band filters fitting 
to geometrical features. Here, we use a following 
simple algorithm to select the filter characteristic. 
Figure 11 shows the principle of the adaptive filter. 
The proposed method compare with the signal 
level's mean of extent (A) and extent (B). If the 
difference of (A) and (B) is large, the wide band 
filter applied. If the difference is small, the narrow 
band filter is applied. 
  
If |A T B| > a, the wide band filter 
1s applied. 
If |A 2 B| <a, the narrow band 
filter 1s applied. 
Fig.11 The principle of the adaptive filter. 
3.RESULTS 
The contour shown in Fig.12 is a result that is 
applied an adaptive filter for noisy image. We can 
measure more accuracy height information owing 
to this method. 
Figure 13 shows error characteristic against the 
filtering parameters of LPF. The adaptive filter 
can obtain more accurate height compared to the 
fixed parameter filter with most suitable y. As 
shown in fig.14, proposed method can obtain more 
accurate data for the image corresponding to 
30dB(S/N). 
Here, we show the analysis of the imaging angle. 
Figure 15 shows a central cross section of the 
results of 3-D analysis in the imaging angle 
parameter, B/H. This shows the restored data by 
the correlation analysis. If the imaging angle take 
a large value, the restored height data can take 
small steps. However, it is shown that the large 
error appears in this case. This is because of the 
increment of the imaging angle causes 
appearances of the  unseeing region and 
deformations of the imaging objects. 
Figure 16 is the result of the restoring error 
against the imaging angle parameter, B/H. In 
Fig.16, we also give the result using other two 
models and show their central cross sections in 
Fig.17. They are models which emphasize and 
attenuate steepness of the original normal terrain 
model, respectively. 
From Fig.16, The most suitable angle exists each 
model. In the normal terrain model, the most 
suitable angle is 0.2. If the imaging angle decrease, 
it is difficult to measure height information 
accurately and the restoring error increase. 
Because step intervals of measured height can not 
represent finely. On the other hand, if imaging 
angle increase, the restoring errors also increase 
because of influence of unseeing region at steep 
topography. Therefore, unless unseeing region 
appears, the B/H had better take large value. 
  
Fig.12 The result of the adaptive filtering. 
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 313