Ackermann: 
Grün: 
Welch (USA): 
Ackermann: 
  
were trying to get better correlation than could be 
done under normal stereocomparator observing. The 
outcome of this was that it did produce a signifi- 
cantly better answer but it was so slow, so expens- 
ive and so impractical that it was never adopted. 
I mention this now because propably the automatic 
image correlation is also extracting the latent 
potential in the photographic image and in effect 
it is now becoming a practical propositon, which 
it wasn't then. 
Thank you for this comment. These were the famous 
Eden blobs, isn't it. At that time still most of 
the work was done in analog instruments the pre- 
cision of which was moderate. Therefore that kind 
of point transfer did not really improve aerial 
triangulation. However, in the meantime, we could 
show that transfer of natural points - not marking 
them - is really capable of very high precision. 
The results are very close to those by signalized 
points and probably very close to what we get now 
with image correlation. As you pointed out the 
real trouble with natural transfer points is the 
administrative problem. You have to describe these 
points and make sure that they are identified in 
all other photographs again. So, in a practical way 
it was not a very handy system, and that is why 
nobody really applied it, at least not regularly. 
Any more questions, please. 
I would appreciate your comments on the impact of 
pixel resolution. What about storing the large 
number of pixels that would be involved. 
To your first question: It can be concluded from 
experience and from theoretical investigations 
that the 20 microns pixel size, which we use at 
the moment, is quite well adapted to the resolution 
of present day aerial photographs. Smaller pixel 
size would not improve the match because the resol- 
ution isn't there. There are some exceptions, 
however. For instance, in order to catch signalized 
points of only 3 pixel diameter you need smaller 
pixel size. At the moment we do not experiment 
with different pixel sizes. 20 microns seem ap- 
propriate for the time being. 
To your second question: We have not really given 
much thought to the problem of storing efficiently 
the digital arrays. But it does not seem to be 
much of a problem. First, vou do not store all 
patches of a block together. Second, it is still 
an open question whether you need storage pixel by 
pixel or whether some efficient data reduction can 
be applied. It also depends on which computer 
facilities you have. I think, it is very feasible 
to have an automated digital system. Certainly 
54