In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
Figure 7. Segmentation result of large buildings obtained through trial-and-error parameter selection, operating time: 2 hours (pan-
sharpened QuichBird MS, Fredericton)
Segmentation of other objects was also tested. The trial-and-
error approach needs 2 to 6 hours to reach acceptable
segmentation results, whereas FdSP optimizer just needs 30 to
60 minutes. If the 90% of the time for manual input and output
is reduced through software integration, FdSP optimizer will
just need less than 10 minutes to select optimal segmentation
parameters for large complex objects.
5. CONCLUSIONS
A Fuzzy-based Segmentation Parameter (FbSP) optimizer was
developed to improve the efficiency of segmentation parameter
selection and accuracy of object segmentation for eCognition.
The FbSP optimizer can be trained using initially segmented
sub segments and the corresponding targeted object of interest.
The FbSP optimizer can then find the optimal segmentation
parameters for the target object, through fuzzy logical analyses
of the target object and its sub objects.
Experiments with QuickBird and Ikonos pan-sharpened MS
images demonstrated that the FbSP optimizer can effectively
find optimal segmentation parameters for objects of interest
within 30 to 60 minutes under the current software
implementation condition. If the current manual and iterative
input and output of feature information between FbSP
optimizer and eCognition is reduced through software
integration, FbSP optimizer will just need a few minutes to find
optimal segmentation parameters for an object of interest. In
contrary, 2 to 6 hours are usually needed for an experienced
operator to find proper segmentation parameters through trial
and error. The proposed supervised approach to automated
determination of optimal segmentation parameters has
demonstrated its superior advantage in speeding up the
segmentation parameter selection and improving the
segmentation quality. It exhibits the potential to boost
segmentation techniques from current trial-and-error stage into
the next stage—semi-automated or automated process.
Further tests with other remote sensing images will be
conducted. The FbSP optimizer will be further improved.
6. ACKNOWLEDGMENTS
This research was supported by the Discovery Research Grants
Program of NSERC (Natural Science and Engineering
Research Council of Canada), Canada Research Chairs
Program, and the Department of National Defence (Canada).
The original QuickBird and Ikonos satellite images were
provided by CFB Gagetown and the City of Fredericton, NB,
Canada.
7. REFERENCES
Baatz, M. and A. Schape, 2000, Multiresolution Segmentation -
An Optimization Approach for High Quality Multi-Scale
Image Segmentation. Angewandte Geographische
Informationsverarbeitung XII, Ed. J. Strobl et al. AGIT
Symposium, Salzburg, Germany, 2000. pp. 12-23.