ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision", Graz, 2002 
  
noise, the two subimages were convolved by a moving 
average filter. The process started by point features 
extraction using Moravec operator (Moravec, 1977), see Fig. 
(2), and then followed by constructing geometric invariant 
features as described in section 2. The two image features 
were paired according to equations (5), (8), (9), and (10). The 
results of pairing were encoded in the relevant parameter 
space as depicted in Fig. (3). The expected registration 
parameters were recovered by searching for the peak value in 
the parameter space. The locus of the peak indicates the 
values of the registration parameters and its peak height 
indicates the number of matched points. Matched points were 
recovered by backtracking the process, as show in Fig. (4). 
Table (1) shows the number of detected and matched points 
between the two images. The matched points are combined 
in a single least squares adjustment, and Table (2) shows the 
results. The adjusted parameters were used to resample the 
second image (SPOT 1991) to the space of the first image 
(SPOT 1987) and Fig. (5) shows the results of resampling as 
image mosaic. Bilinear transformation is used as an 
interpolation method in the resampling process. 
  
SPOT 1987 
SPOT 1991 
Figure 1: Two SPOT subimages, taken at different time (1987 and 1991), over the Hanford Reservation in Washington State, USA. 
  
SPOT 1987 
Figure 2: Shows the results of point features extraction using Moravec operator. 
  
i 
SPOT 1991