Automatic DEM Generation From Satellite Image
2. DEM generation method
This section briefly introduces the DEM generation method used in this paper. The procedure of
DEM generation is shown in figure 1. We firstly generate geometric sensor model for those
images with ancillary data, such as SPOT 4 & 5. Because some high resolution satellite images
only provide RPC (rational polynomial coefficients) as geometric sensor model, in order to keep
identical with this sensor model, we chose RPC as the final geometric sensor model in the DEM
generation procedure. So we must generate RPC for those images with ancillary data. Next step,
we should use ground control points or image tie points to refine these geometric sensor models.
After refinement, these geometric sensor models can be used in later processing.
Before image matching, we do some processing on the original images. In most bright and dark
areas, such as mountain area, the lack of features directly leads wrong matching results. We
firstly use Wallis filter (Wallis, 1970) for an adaptive nonlinear local contrast enhancement. This
enhancement produces good local contrast throughout the image, while reducing the overall
contrast between bright and dark areas.
Figure 1, flow chart
After Wallis filtering, the gradient magnitude images are computed. To reduce weak edges due
to noise, all gradients with a magnitude less than a threshold T are set equal to T. The threshold
is selected as a function of the mean and the standard deviation of the gradient magnitude image
(in our program, T=mean-standard deviation). The same function is used for both images to
ensure equal treatment. The threshold should not be too high otherwise (a) useful texture is
deleted, and (b) the edges are broken and significant differences between the two images occur
due to different edge strength. This approach eliminates noise but also low texture which is
however not very likely to lead to accurate matching results. (Emmanuel P. Baltsavias, Dirk
Stallman, 1993).
After the generation of image pyramid, we do image matching. For each pixel, we do not give
its final matching result firstly but a group of matching results with corresponding correlation
value. After all pixels have been matched, we do not transfer the matching result to the next
pyramid floor immediately but a relaxation procedure (Li Zhang, Maria Pateraki, Emmanuel
Baltsavias, 2002). With this relaxation procedure, for each pixel, we select a matching result
from the candidate group which is most compatible to its neighbourhoods as the final matching
result. After the relaxation procedure is completed, we transfer the matching result to the next
pyramid floor and start the next round of image matching. The whole procedure is a iteration
procedure till the final pyramid floor.
Finally the matching result was interpolated into DEM whose resolution was defined by user
before. We do not need any post processing on this output DEM. The whole procedure is fully
automatic.
