International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
Thus, the results of visual compared qualities results of the
‘individual segmentation programs are reinforced. Only the
segmentations by SPRING as well as eCognition 2.1 and 3.0
have reached good overall results. These programs leading to
the slightest differences to the reference areas at all factors
investigated. Likewise InfoPack and the ,Minimum Entropy
Approach' yielded to an acceptable quality, but InfoPack tends
to over-segmentation and the ,Minimum Entropy Approach’
has some processing problems as stated above. The results of
the three remaining programs did not reach this quality. They
probably failed due to the high complexity of high resolution
remote sensing imagery. Often a strong faulty or over-
segmentation is the consequence. Furthermore, the grade of
conformity with the reference objects is only slight. Indeed it
has to be reemphasised, that some of the approaches have not
primarily been developed for (optical) remote sensing image
analysis.
5. CONCLUSIONS
Due to the dissimilitude of the software implementations the
segmentation results are naturally varying. It was shown, that
best results have been calculated using the commercial software
packages — eCognition and InfoPACK. The only exception is
the freeware SPRING, but with the disadvantages of a higher
operating expense and a worse handling. However, the use of
InfoPACK leads to more over-segmented results. Another
algorithm with a high potential is the ‘Minimum Entropy
Approach to Adaptive Image Polygonization', but there was
also an over-segmentation. The results of the other programs
were not satisfying user's demands.
[mage segmentation has become essential for high resolution
remote sensing imagery. The further development of first
promising segmentation approaches offers a lot of potentials to
make remote sensing image analysis more accurate as well as
more efficient. The use of texture information for segmentation
could improve the results. Indeed at the moment only
InfoPACK provides this option, which was not used for this
evaluation. Increasing combinations, for instance with
algorithms of feature extraction, edge-oriented or model-based
segmentation should be aspired for the improvement of
segmentation quality.
Segmentation algorithms respond often very sensitively in the
case of negligible variations, like slight parameter chances, the
order of segmentation hierarchical approaches or the image data
itself (image size, bit depth, etc.). Thus, the user is confronted
with a high degree of freedom, which should be minimised. For
instance, when selecting parameters by the trial-and-error
method the results are highly influenced by subjectivity. The
integration of instruments for evaluation of segmentation
quality appears desirable.
[n future additional segmentation programs will be evaluated,
for instance the image processing systems HALCON and
IMPACT. Moreover, this more qualitative evaluation will be
added by a quantitative comparison using the software SEQ-
Tool (Delphi IMM GmbH, 2003). This tool compares the
identicalness of polygon outlines (segmented vs. reference).
REFERENCES
Baatz, M. & Schäpe, A, 2000: Multiresolution Segmentation -
an optimization approach for high quality multi-scale image
segmentation. In: Strobl, J. et al. (eds.) Angewandie
Geographische | Informationsverarbeitung XII. Wichmann,
Heidelberg, pp. 12-23.
Bins, L. S.; Fonseca, L. M. G.; Erthal, G. J. & Ii, F. À. M.
(1996): Satellite imagery segmentation: a region growing
approach. Proceedings of VIII Brazilian Remote Sensing
Symposium, Salvador, Bahia: 4 p.
Cook, R.; McConnell; I., Stewart, D. & Oliver, C., 199%:
Segmentation and simulated annealing. In: Franceschetti, G. et
al. (eds.): Microwave Sensing and Synthetic Aperture Radar.
Proc. SPIE 2958, pp. 30-35.
Delphi IMM GmbH, 2004: Bestimmung der Segmentierungs-
qualität bei objektorientierten Bildanalyseverfahren mit SEQ.
http://www.delphi-imm.de/neu/ ? Fernerkundung > Software >
SEQ (in German only, accessed 26 Apr. 2003).
Haralick, R. & Shapiro, L., 1985: Image segmentation
techniques. Computer Vision, Graphics, and Image Processing,
vol. 29, pp 100-132.
Hermes, L. & Buhmann, J. M., 2001: A New A daptive
Algorithm for the Polygonization of Noisy Imagery. Technical
Report IAI-TR-2001-3, Dept. of Computer Science Ill,
University of Bonn.
Kettig, R. L. & Landgrebe, D. A. 1976: Classification of
Multispectral Image Data by Extraction and Classification of
Homogeneous Objects. [EEE Transactions on Geoscience
Electronics, Vol. GE-14, No. 1, pp. 19-26.
Meinel, G.. Neubert, M. & Reder, J., 2001: The potential use of
very high resolution satellite data for urban areas — First
experiences with IKONOS data, their classification and
application in urban planning and environmental monitoring
In: Jürgens, C. (ed.): Remote sensing of urban areas.
Regensburger Geographische Schriften 35, pp. 196-205.
Pal, N°R. & Pal, S.K., 1993: À review on image segmentation
techniques. Pattern Recognition, vol. 26, pp. 1277-1294.
Ruefenacht, B., Vanderzanden; D., Morrison, M. & Golden, M,
2002: New Technique for Segmenting Images. Technical
documentation.
von Ferber, C. & Wörgötter, F., 2000: Cluster update algorithm
and recognition. Phys. Rev. E 62, Nr. 2, Part À, pp. 1461-1464.
ACKNOWLEDGEMENTS
The authors wish to thank the German Research Foundation for
claiming the project “Use and application of high resolution
satellite imagery for spatial planning” (Me 1592/1-2). Ft
processing the different segmentations we thank Ms. Prictzsc
(Infoterra), Mr. Oliver (InfoSAR), Mr. von Ferber (University
of Freiburg) and Mr. Hermes (University of Bonn).
1102
KEY
ABS
Lack
scale
explc
Suppx
prepa
In th
were
propc
meth
mode
result
other
Miner
small
topog
collec
produ
smalle
Miner
are vi
Inforn
retreiv
diffère
as: lar
non-e»
manag
Storing
integre
explor.
In this
mappi
conver
mappir
Copper
finally
In min
particu
After a
