S.A.B. Kim 
2 DATA 
AL Each pair of SPOT stereo-images cover an area of 60 km x 60 km. Stereo-matching is an area-based one using the 
epipolar correlation criterion (Kim, 2000). About 20 GPS measurements are used as ground control points. To save 
computing time, stereo-matching is performed at every 5 pixel, thus the output resolution is 50 m. The stereo-match 
results are transformed into elevation using a camera model by Orun and Natarajan (1994). Gaussian interpolator of 
Kim et al. (1999) produces a DEM shown in Fig. 1a. 
  
nted, 
s that 
iders, 
Even 
s are 
buted 
r not 
ation 
0 km 
ints. 
  
  
  
  
  
Fig. 1 Limitations of conventional interpolation methods (a) generated by Gaussian interpolator without intelligent 
ation methods and (b) PCI DEM over a 60 km x 60 km area in Seoul/KyungKi region, S. Korea. The coast is on the left. 
ning, 
alues 
ating | 3 CENTER-OF-GRAVITY (COG) AND EMPTY-CENTER-INDEX (ECI) 
urac | 
y COG and ECI are used to remove unrealistic outstretch of elevation along an image boundary. Interpolated values 
Foley whose COG or ECI fail to meet threshold values are removed. 
tered 
0 km 
ation 
POT 
villa PF 
Iging oam e Y x 
along | fis 3 
| 
  
'oast; 
lified 
total 
idex) 
lance . ARN % : 
ed of [p iy) 
I and 
g the 
ation, 
This 
ently Fig. 2 Illustration of ‘center-of-gravity’ (COG, left) Fig. 3 Distribution of COG (dotted) and ECI (solid) computed 
their and ‘empty-center-index’ (ECI, right). ECI may be over a 35 km x 30 km area. The total number of COG (and 
regarded as a 1D version of COG (1D being in the  ECI) samples is 471,000. 
radial direction). The shaded area is where stereo- 
match succeeded, while the white area is not. The 
outer circle indicates an interpolation radius. 
  
  
  
  
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 497