using the 
periments 
unknown 
0 the next 
n for the 
pectively. 
error-free 
TM. This 
nbert and 
error-free 
| which is 
yerformed 
with the 
1-Seeliger 
as initial 
rong and 
has been 
5%, 10%, 
rrect one 
ages. 
e height 
ments can 
ite height 
ted more 
e element 
tions (10) 
ht offset 
shifted in 
ey value 
ements to 
eriments, 
al height 
ee albedo 
Lommel- 
nces AZ 
It of the 
[ AZ/h =~ 
luring the 
ert case, 
correctly; 
cal plane 
e correct, 
lane are 
Figure 7 
ong with 
nstructed 
direction. 
ITERATION NO. 
3 
es 
css 
    
  
A 
^ 
0 
SB 
SZ 
SAN 
eth 
OO0 
XXX 
v 
0 
^ 
  
  
QU 
QN 
Q 
Q 
QOO 
QV. 
       
      
X 
o0 
SX 
iet 
o 
3 
SS 
á 
Figure 7: Result of the surface reconstruction with the Lambert model (A4 known, error-free; horizontal plane 
as initial DTM) 
LDHERATIONSNO..:-$ 
  
Figure 8: Result of the surface reconstruction with the Lommel-Seeliger model (wrong, error-free Ag; 
horizontal plane as initial DTM) 
ITERATION NO. “50 
Figure 9: Result of the surface reconstruction with the Lommel-Seeliger model, if a Lambert surface is 
introduced 
independent of viewpoint, the three images of the 
surface are equal, except for different perspective 
distortions. Consequently, the additional images do 
not contribute additional information to the 
estimation of the surface inclinations perpendicular to 
the mentioned plane - this is in line with the general 
indeterminability of classical SFS (see introduction). 
- The reconstruction of a surface with error-free 
heights and a wrong, error-free albedo which is 5% 
lower than the correct one causes the whole surface 
to be tilted towards the illumination direction, 
because all grey values which are computed from the 
initial values are a constant factor smaller than the 
image grey values. This effect is nearly independent 
of the surface topography, as can be seen from the 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
differences between reference DTM and 
reconstructed DTM (see figure 8); the height 
differences between neighbouring DTM heights are 
nearly constant. All grey value differences are 
interpreted as resulting from an incorrect angle 
between local surface normal and illumination 
direction s. 
The reconstruction of a Lambertian surface by the 
Lommel-Seeliger approach causes the whole surface 
to be tilted extremely away from the viewing 
direction, since large grey values in the Lommel- 
Seeliger law are directly related to large emittance 
angles e (see figure 9). In contrast to the simulation 
with wrong, error-free albedo, there is also an 
influence of surface topography on the differences