Documente Academic
Documente Profesional
Documente Cultură
1, JANUARY 2006
Abstract—The classification of very high-resolution remotely 1) Feature extraction: This is based on the construction of
sensed images from urban areas is addressed. Previous studies a derivative morphological profile, which characterizes
have shown the interest of exploiting the local geometrical in- each pixel, both in terms of intensity and in terms of local
formation of each pixel to improve the classification. This is geometry. This step is described in Section II.
performed using the derivative morphological profile (DMP) ob-
tained with a granulometric approach, using opening and closing
2) Classification: This step was initially based on a neural
operators. For each pixel, this DMP constitutes the feature vector network. In this letter, we propose to use a fuzzy possi-
on which the classification is based. In this letter, we present an bilistic model instead. This is presented in Section III.
interpretation of the DMP in terms of a fuzzy measurement of The results obtained on IKONOS remote sensing data are
the characteristic size and contrast of each structure. This fuzzy presented in Section IV. Fig. 4 presents one original panchro-
measure can be compared to predefined possibility distributions to matic image from Reykjavik, Iceland (1-m resolution, size
derive a membership degree for a set of given classes. The decision
is taken by selecting the class with the highest membership degree.
975 639). Six classes of interest are defined: large buildings,
This model is illustrated and validated in a classification problem houses, open areas, large roads, streets, and shadows (see [2]
using IKONOS images. for more details). The proposed method is positively compared
Index Terms—Classification, fuzzy sets, granulometry, mathe-
to the previously published method.
matical morphology, possibility distribution, urban area, very high
resolution. II. FEATURE EXTRACTION AND GRANULOMETRY
Granulometriesarepopularandpowerfultoolsderivedfromthe
I. INTRODUCTION mathematicalmorphologytheory.Theyareclassicallyusedforthe
analysisofthesizedistributionofparticlesinanimage.Theycanbe
(a) (b)
Fig. 3. Possibility distributions modeling (a) low and (b) high contrast.
class (6)
Note that the idea of jointly processing the size and the contrast
of the different features in an image has also been addressed in
[13] for the analysis of oriented patterns such as fingerprints or
fiber structures.
IV. RESULTS
In our experiments, we used the following parameters: eight
closings and eight openings were computed with increasing
structuring element sizes (the increment is three pixels). This is
a tradeoff to obtain both a feature vector with a reasonable size (c) (d)
and SE sizes ranging from small (3) to rather large Fig. 4. (a) Original Ikonos image. (b) Manually labeled ground truth. (c)
. The setting of these parameters must ensure that Classification result using the neural network. (d) Classification result using
all the structures of interest fall within this range. The increment the proposed fuzzy approach.
should be fine enough to ensure the separation between the
classes of interest, consistently with the expert’s definition of
TABLE I
the corresponding distributions of possibility. As previously CONFUSION MATRIX—NEURAL CLASSIFIER—OVERALL ACC. 40.3%
mentioned, the structuring elements are discs. Fig. 4(b) presents
the ground truth obtained by a manual labeling. It is especially
large to ensure a fair and significant evaluation. For visual
inspection, Fig. 4(c) presents the classification results obtained
with the neural network (algorithm presented in [2]). Fig. 4(d)
presents the results obtained with the proposed fuzzy possi-
bilistic model. The following lookup table is used: shadows are
in brown, large buildings in red, houses in gray, large roads
in greenish blue, streets in blue, and open areas in green. For
a quantitative evaluation, Tables I and II, respectively, present
the confusion matrices obtained with the two methods for the
six classes. It is interesting to see that the overall accuracy
increases from 40.3% with the neural network to 52.1% with
the proposed fuzzy approach. However, one should also note
that the neural network outperforms the proposed method in
terms of accuracies for the streets class. postfiltering could increase these accuracies. Although the rep-
It should be underlined that the numerical results (Tables I resentation of the data in terms of DMP has proven its poten-
and II) should only be considered in a relative way, in order tial, it also has some intrinsic limitations (accuracy results are
to compare the performances of the methods. Some pre- and still far below 100%, meaning that some further developments
44 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 3, NO. 1, JANUARY 2006
REFERENCES
[1] M. Pesaresi and J. A. Benediktsson, “A new approach for the morpho-
are still needed to solve the problem of automatic classifica- logical segmentation of high resolution satellite imagery,” IEEE Trans.
tion of very high-resolution images from urban areas). In par- Geosci. Remote Sens., vol. 39, no. 2, pp. 309–320, Feb. 2001.
ticular, the DMP only provides information regarding the size [2] J. A. Benediktsson, M. Pesaresi, and K. Arnason, “Classification and
and the contrast of the objects within the image. To achieve feature extraction for remote sensing images from urban areas based on
morphological transformations,” IEEE Trans. Geosci. Remote Sens., vol.
a more accurate classification, additional information may be 41, no. 9, pp. 1940–1949, Sep. 2003.
needed, especially in the context of complex natural images. [3] R. Gonzalez and R. Woods, Digital Image Processing, 2nd ed: Prentice
Various methods incorporating contextual, structural, or percep- Hall, 2002.
[4] J. Serra, Mathematical Morphology, Theoretical Advances. Orlando,
tual information have been proposed [14]–[16]. However, this is FL: Academic, 1988, vol. 2.
beyond the scope of this letter whose aim is to propose and eval- [5] P. Soille, Morphological Image Analysis, Principles and Applications,
uate an alternative way of handling the DMP and extract valu- 2nd ed. Berlin, Germany: Springer-Verlag, 2003.
able information. [6] P. Soille and M. Pesaresi, “Advances in mathematical morphology ap-
plied to geoscience and remote sensing,” IEEE Trans. Geosci. Remote
Sens., vol. 40, no. 9, pp. 2042–2055, Sep. 2002.
V. CONCLUSION [7] J. Crespo, J. Serra, and R. Schafer, “Theoretical aspects of morpholog-
ical filters by reconstruction,” Signal Process., vol. 47, pp. 201–225,
The contribution of this letter lies in the two following items. 1995.
• Fuzzy interpretation of the DMP: Based on its basic defi- [8] L. Zadeh, “Fuzzy sets,” Inf. Control, vol. 8, pp. 338–353, 1965.
[9] , “Fuzzy sets as the basis for a theory of possibility,” Fuzzy Sets
nition, the differential morphological profile is interpreted Syst., vol. 1, pp. 3–28, 1978.
as a fuzzy measurement of the characteristic size of the [10] D. Dubois and H. Prade, Possibility Theory. New York: Plenum, 1988.
objects. It also provides information on the contrast of the [11] Z. Wang and G. Klir, Fuzzy Measure Theory. New York: Plenum,
structures. 1991.
[12] I. Bloch, “Information combination operators for data fusion: A compar-
• Possibilistic model: We show how the expert’s knowledge ative review with classification,” IEEE Trans. Syst., Man, Cybern., vol.
regarding the size and the contrast of the structures of in- 26, no. 1, pp. 52–67, Jan. 1996.
terest can be modeled as possibility distributions. [13] P. Soille and H. Talbot, “Directional morphological filtering,” IEEE
Trans. Pattern Anal. Mach. Intell., vol. 23, no. 11, pp. 1313–1329, Nov.
For classification purposes, the measured DMPs (interpreted as 2001.
fuzzy sets) are compared with the previously designed possi- [14] A. Shackelford and C. Davis, “A combined fuzzy pixel-based and ob-
bility distributions. For each pixel, this provides a membership ject-based approach for classification of high resolution multispectral
data over urban areas,” IEEE Trans. Geosci. Remote Sens., vol. 41, no.
degree to each of the classes. The final decision is taken by 10, pp. 2354–2363, Oct. 2003.
selecting the class with the highest resulting membership de- [15] P. Gamba, J. A. Benediktsson, and G. Wilkinson, “Foreword to the spe-
gree. This proposed classification method has several advan- cial issue on urban remote sensing by satellite,” IEEE Trans. Geosci.
tages, which are summarized in the three following items. Remote Sens., vol. 41, no. 9, pp. 1903–1906, Sep. 2003.
[16] C. Unsalan and K. Boyer, “A system to detect houses and residential
• Simplicity: The trapezoidal model used for each possi- street network in multispectral satellite images,” Comput. Vis. Image Un-
bility distribution is simple and uses few parameters. Being derstanding, 2005, to be published.
fuzzy, its settings are not too sensitive. Fitting intuition, [17] M. Fauvel, J. Chanussot, and J. A. Benediktsson, “Fusion of methods for
the classification of remote sensing images from urban areas,” presented
the model does not require any manual ground truth nor at the IGARSS, Seoul, Korea, 2005.
training. [18] J. A. Benediktsson, J. A. Palmason, and J. R. Sveinsson, “Classification
• Flexibility: In the proposed method, it is very easy to add of hyperspectral data from urban areas based on extended morphological
a new class of interest without restarting the whole clas- profiles,” IEEE Trans. Geosci. Remote Sens., vol. 43, no. 3, pp. 480–491,
Mar. 2005.
sification process. There is only one possibility distribu- [19] A. Plaza, P. Martinez, J. Plaza, and R. Perez, “Dimensionality reduc-
tion to define and the corresponding membership degree tion and classification of hyperspectral image data using sequences of
to compute. extended morphological transformations,” IEEE Trans. Geosci. Remote
Sens., vol. 43, no. 3, pp. 466–479, Mar. 2005.
• Complementarity: Finally, it is interesting to note that the [20] J. Chanussot, G. Mauris, and P. Lambert, “Fuzzy fusion techniques for
proposed algorithm and the previously published one [2] linear features detection in multitemporal sar images,” IEEE Trans.
provide complementary results. Therefore, there is a poten- Geosci. Remote Sens., vol. 37, no. 3, pp. 1292–1305, May 1999.