The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
99
Although the segments obtained will tend to stay compact (con
trary to regions gained e.g. by spectral clustering where no spatial
information is used) they do not have to be connected. The root of
a subtree at a high level covers a large area within the image and
can thus connect regions, which are neither close to each other
nor connected, but have similiar spectral properties. Another im
portant feature is that the number of segments is exclusivly based
on the given data. Neither an exact number has to be set, nor a
maximum number, because the number of segments is a direct
result of the algorithm.
As the proposed method was designed to segment homogeneous
areas, one of its limits is shown in Figure 4. It is not able to con
sider a heterogeneous image area like regions with strong tex
ture as one segment. Such a region (e.g. cities or forests with
great fluctuations in height and/or changes in backscatter proper
ties due to different vegetation) will be segmented in many small
regions. Future work will include the analysis of all regions in
a more rigorous way than the above mentioned simple merging
algorithm to overcome this disadvantage.
Figure 4: top: original; middle: segmentation; bottom: segment
borders
REFERENCES
Lee, J.-S. and Gruñes, M.R. and Kwok, R., 1994. Classifica
tion of multilook polarimetric SAR imagery based on complex
Wishart distribution. Int. J. Remote Sensing, Volume 15, No.11,
pp. 2299-2312.
Ferro-Famil, L. and Pottier, E. and Lee, J.-S., 2001. Unsuper
vised classification of multifrequency and fullypolarimetric SAR
images based on the H/A/Alpha-Wishart classifier. IEEE Trans
actions on Geoscience and Remote Sensing, Volume 39, Issue 11,
pp. 2332-2342.
Anfinsen, S. N. and Jenssen, R. and Eltoft, T., 2007. Spec
tral Clustering of Polarimetric SAR Data with Wishart-Derived
Distance Measures. Proc. Inti. Workshop on Science and Ap
plications of SAR Polarimetry and Polarimetric Interferometry
(POLinSAR2007).
Hänsch, R. and Jäger, M. and Hellwich, O., 2008. Clustering by
deterministic annealing and Wishart based distance measures for
fully-polarimetric SAR-data. Proc. EUSAR2008.
Reigber, A. and Jäger, M. and Neumann, M. and Ferro-Famil, L.,
2007. Polarimetric Fuzzy K-Means Classification with Consider
ation of Spatial Context. Proceedings of POLINSAR’07.
Hong, T.H. and Rosenfeld, A., 1984. Compact Region Extrac
tion Using Weighted Pixel Linking in a Pyramid. IEEE Transac
tions On Pattern Analysis and Machine Intelligence, Vol.PAMI-
6,No.2, pp. 222-229.