International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
cloud shadows. The DS fusion removes this kind of error by.de-
creasing the credibility associated to this decision. The rate of
classification after DS fusion is 63 96 .
The final step is based on an evidential markovian approach.
Each classified pixel is studied with a spatial context and its own
mass function after the DS fusion [1]. The algorithm used in
this step is the evidential ICM (Iterative Conditional Mode). The
rate of good classification after this algorithm increase to 66 %
(fig.2.d).
CONCLUSION
The new evidential fusion approach can distinct in a better
way the uncertainty et inaccuracy notion in the mass functions.
The FSEM algorithm compared to others algorithms used in the
DS fusion enables possible to distribute a more realistic density
for each simple or composed hypothesis. We have seen that the
mass function initialization proposed in this article ease the fu-
sion process. The mass function are no more estimated using
an empirical approach, so the algorithm is completely automatic
and we do not need any a priori information about the data.
Application to remote sensing multispectral images with
some ecological indices and auxiliary data related to slope in-
formation give a better classification result on the final decision-
making. Redundant and heterogeneity information decrease
some ambiguities related to a lack of data and some artefacts.
Thus, the DS fusion developed method improves the result in
comparaison with the result of LANDSAT TM SEM classifica-
tion.
REFERENCES
[1] Bendjebbour, A., Delignon, Y., Fouque, L., Samson, V. and Pieczynski, W.
(2001) Multisensor Image Segmentation Using Dempster-Shafer Fusion in
Markov Fields Context, IEEE Trans. On Geoscience and Remote Sensing,
vol. 39, p. 1789-1798.
[2] Bentabet, L., Zhu, Y. M., Kaftdandjian, V., Babot, D. and Rombaut,
M.(2000) Use of fuzzy clustering for determining mass functions in
Dempster-Shafer theory, presented at Signal Processing Proceedings,
WCCC-ICSP 2000. 5th International Conference on.
[3] Bloch, I. and Maitre, H. (1994) Fusion de Données en Traitement d'Images :
modeles d'information et décisions, Traitement du Signal, vol. 6, p. 1811-
1823.
[4] Bracker, H. (1996) Utilisation de la théorie de dempster/Shafer pour la clas-
sification d'images satellitaires à l'aide de données multi-sources et multi-
temporelles. : Rennes I, p. 178.
[5] Caillol, H., A., H. and Pieczynski, W. (1993) Fuzzy random fields and unsu-
pervised image segmentation, IEEE Transaction on Geoscience and Remote
Sensing, vol. 31, p. 801-810.
[6] Dempster, A. P. (1968) A generalisation of Bayesian inference, Journal of
the Royal Statistical Society, p. 205-247.
[7] Germain, M., Voorons, M., Boucher, J.-M. and Benie, G. B.(2002) Fuzzy
statistical classification method for multiband image, presented at Procee-
dings of the Fifth International Conference on Information Fusion, Anna-
polis, Maryland, USA.
[8] Lee, T., Richards, J. A. et Swain, P. H. (1987) Probabilistic and Eviden-
tial Approches for Multisource data Analysis, IEEE Transactions on Geos-
cience and Remote Sensing, vol. 25, p. 283-293.
[9] Peddle, D. R. (1993) An Empirical Comparaison of Evidential reasoning,
Linear Discriminant Analysis and Maximum Likelihood Algorithms for Al-
pine Land Cover Classification, Canadian Journal of Remote Sensing, vol.
19, p. 31-44.
[10] Salzenstein, F. and Pieczynski, W. (1997) Parameter estimation in hidden
fuzzy Markov random fields and image segmentation, Graphical Models
and Image Processing, vol. 59, p. 205-220.
[11] Shafer, G. (1976) A Mathematical "Theory of Evidence", vol. Princeton,
NJ.
[12] Verikas, A., Malmqvist, K. and Bacauskiene, M.(2000) Combining Neural
Networks, Fuzzy Sets, and Evidence Theory Based Approaches for Ana-
lysing Colour Images, presented at Neural Networks, 2000. IJCNN 2000,
Proceedings of the IEEE-INNS-ENNS International Joint Conference on.
[13] Zadeh, L. A. (1965) Fuzzy sets, Information and Control, vol. 8, p. 338-
352.
1266
