-25- 
(Bernstein /8/). Both institutions operate their direct or inverse routines by aid of correction 
polynomials. 
A program package involving indirect transfer using collinearity equations and a direct transfer 
using correction polynomials has been developed at the Institute of Photogrammetry, Technical 
University of Hannover /50/, /78/. 
  
6. Image Registration 
For the purposes of multitemporal comparisons image data must be registered with respect to 
each other. While a few years ago image correlation over the entire image has been propagated 
for this purpose, one now finds that (until the eventual arrival of special processors) geometric 
differential rectification offers a more time consuming and a more cost saving approach. The cost 
and computer time relation between rectification and digital image correlation has been esti- 
mated at LARS at 1 : 3. In the absence of efficient electronic-analog or hybrid special correlators 
which could be built in principle for images or for digital data streams from tape rectification 
to a geometric base appears to be the most suitable approach. On the other hand the rectification 
process becomes tedious and inaccurate because of the necessity to measure image coordinates 
of ground control points. For that reason digital imag correlation may be adopted for that pur- 
pose only in all cases where a repetitive coverage over a known area (such as the U.S.A. ) is 
expected. 
For that reason digital image correlation programs have been developed at Jet Propulsion Labora- 
tories (JPL) and at IBM. 
The JPL-Program is an interactive one, where the search for apoint may be minimised by moving 
a cursor to the approximate point locations on 2 TV screens. Then the image correlation routine 
is initiated for 64 x 64 pixel squares. The algorithm used consits of the multiplication of the 
two 2-D Fourier transforms of the images which have been generated by a special Fourier transform 
hardware processor. 
The I.B.M. - program intensity normalizes both samples to be correlated and forms their respec- 
tive differences. If the difference sum is larger than a certain Hueshold a new sample (shifted 
in x or y is compared. A minimum difference sum will be found for the approximate match. In this 
area 5 x 5 correlation coefficients are calculated. A surface is formed from these and the max- 
imum is determined, resulting in an accuracy of match up to 1/3 the pixel size. 
7. Conclusions 
A summary of the approach, the state of the art and the trends of geometric restitution for remote 
sensing imagery has been attempted. Considering the short time since the introduction of remote 
sensors into photogrammetry one cannot help but wonder about the rapid progress, which in part 
has significantly come about from groups outside the (classical) field of photogrammetry. The 
conclusion of this paper must be that these groups should find more representation in the Inter- 
national Society of Photogrammetry. The working group III - 1 may continue to be a catalyst for 
further progress.