352 
By assuming that all pixels of a segment contribute equally to a 
match (which holds only statistically, but not for individual seg- 
ments), a target segment can be regarded as a rectangular sampling 
function. From [2] it follows that fidelity decrases by increasing 
the overlap o: TE, = f(r, 0), where r, 1s the rectangular sampling 
function representing the target, and o is target overlap. 
ba 1 target | 
| | 
i 
  
À target I | target — = 
| 
  
  
  
  
  
  
  
  
  
  
  
  
  
— 
  
  
| 1% 
i 77 7 7 
| 
  
pixeT Co quus er 
Fig. 5: Overlapping target segments. 
If a linear filter is applied for smoothing parallaxes [6], the 
corresponding degradation effect can be represented by: TF, -f(filter). 
The combined degrading effect of incrementing (A) of overlapping 
segments (0o), and of filtering can be represented by the Transfer 
Function TF = TF, . TF 5° TF5» 
ANNEX 6 Fidelity of DTM 
In an ideal case, the fidelity of the DTM would be identical to that 
of the corresponding parallax data, i.e., TF, = TF,- 
Raw DTM data have to be converted and resampled, e.g., from epipolar 
profiles to a uniform grid in a terrestrial coordinate system. The 
corresponding Transfer Function TF« - f(A, r), where ^ is the new 
grid interval and T is a (virtually) rectangular function represen- 
ting a new grid cell. 
If parallax data were not smoothed before (annex 5), a linear 
smoothing filter can be applied to DTM data; hence, TF, - f(filter). 
Fidelity of the resulting DTM is therefore determined by 
For a still acceptable (minimum) fidelity, the corresponding terrain 
resolution (i.e., the smallest resolvable terrain form) can be de- 
termined by means of the TFpTM [6]. 
353