Manfred H. Günzl
improve the segmentation quality slightly. As an example using SCV - uA(7,,,,, P)/ l5s gives a result of e — 0.73 shown
in image 6(f). Where / is the length of the boarder to remove and 4 is Zadehs S-Function used to control the influence
of r,,,,, dependent on the segment size P.
4 CONCLUSIONS
The application of the shape parameter r;,,. in a region growing algorithm pinpointed two mayor problems.
e The decision of a further fusion step needs to be derived out of a combination of radiometric and geometric param-
eters. Even if both of them are sources of evidence they can disturb each other if combined to a merging criterion.
Applied on single look SAR date with a signal to noise ratio of one in most cases the application of a shape param-
eter leads to worse results as shown in image 6(e). In the early stage of the segmentation the correlated speckles
with there compact shapes are increased. Using more complicated operators and weighting parameters to combine
radiometric and geometric parameters can solve this problem. The drawback that prevents the practical application
is that right now there is no way to derive these parameters from a priori knowledge. The better the signal to noise
ratio the less critical the combination of radiometric and geometric parameter is.
e The second principal problem is the application of r,,,. in greedy algorithms. Trying to keep segments as rectangular
as possible during the entire segmentation process sometimes prevents the algorithm from finding large rectangular
objects that need to be assembled out of non compact shapes. Comparisons with simulations of uncorrelated speckle
showed that this is particularly valid when applied to data with correlated speckle.
ACKNOWLEDGMENTS
These investigations are a part of the EMAP project. EMAP is equally financed by the German Aerospace Center (DLR)
and the German Federal Ministry of Agriculture (BML). The test site is located in southern Bavaria between Munich and
the Danube river. Three sequences of 7-8 ERS-2-SAR-SLC images per year, acquired between March and October, have
been kindly provided by the European Space Agency (ESA).
REFERENCES
Beaulieu, J.-M. and Goldberg, M., 1989. Hierarchy in picture segmentation: a stepwise optimization approach. IEEE
Transactions on Pattern Analysis and Machine Intelligence 11(2), pp. 150—163.
Cormen, H. T., Leiserson, C. E. and Rivest, R. L., 1990. Introduction to Algorithms. McGraw-Hill.
Foley, J. D., van Dam, A., Feiner, S. K. and Hughes, J. F., 1997. Computer Graphics — Principles and Practice. 2 edn,
Addison-Wesley Publishing. ISBN: 0-201-84840-6.
Schachter, B., Davis, L. and Rosenfeld, A., 1979. Some experiments in image segmentation by clusterin of local feature
values. Pattern Recognition 11(1), pp. 19-28.
Tilton, J. and Cox, S., 1983. Segmentation of remotely sensed data using parallel region growing. In: Digest of the 1983
International Geoscience and Remote Sensing Symposium — San Francisco, CA, pp. 9.1-9.6.
Tilton, J. and Ramapriyan, H., 1988. Data compression experiments with landsat thematic mapper and nimbus-7 coastal
zone color scanner data. In: Proceedings of the Scientific Data Compression Workshop — Snowbird, UT.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 357