The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
230
in terms of keeping the extent of intensity variation (RMSE and
NMSE) small and the structural similarity (SSIM) with the refer
ence hypercube high. Such elegantly simplified data can be used
instead of the original noisy ones, improving the performance of
the succeeding band selection, feature extraction and classifica
tion procedures, especially the unsupervised ones. The AML,
naturally, provides a simpler space for statistical modeling and
interpretation, by preserving distinguishable data features while
reducing spatial and spectral intensity variation.
Moreover, the compared filtering techniques were applied in re
moving noise from an artificially contaminated hypercube. One
percent of the original hypercube’s pixels were contaminated with
uncorrelated noise and then ADF, ML and AML of scale n=6
were applied. The quantitative measures when comparing results
with the original hypercube are presented in Table 1. The devel
oped AML scores better in all measures approximating success
fully the original hypercube’s intensity and structure. Last but not
least, the AML was tested against watershed’s over-segmentation
problem. In all performed experiments, a reduction of over a 10%
was achieved to the number of the output segments. AML man
aged to decrease the heterogeneity of the initial image (both in
spectral and spatial directions) by merging pixels which belonged
to the same object/class, impelling the sensitive watershed trans
formation to result in fewer output segments.
5 CONCLUSIONS
We have introduced a novel morphological scale space repre
sentation for denoising and simplifying hyperspectral data. Ex
perimental results and performed quantitative evaluation demon
strate that the developed AML can enlarge and create new flat
zones without displacing image contours and can surpass spec
tral spike-like features outperforming anisotropic diffusion filter
ing and standard MLs. The algorithm is relative fast and with
out an optimized coding, can approximately process a hyper
cube of 200x350 pixels with 160 channels in less than a minute
(for every scale n) in an ordinary iPentiumM 2GHz,lGB RAM.
For real-time applications its implementation on a parallel sys
tem is straightforward and furthermore, the algorithm can be ad
justed and do not process the thermal infrared bands, other heav
ily noised or selected ones. The suitable for hyperspectral data
morphological framework, the resulting, in all our experiments,
elegant simplification and the adequate algorithm’s performance
encourage future research. Object-oriented hyperspectral image
analysis, where the multiscale segmentation and classification is
constrained by the developed AML is currently under investiga
tion.
REFERENCES
Benediktsson, J. A., Palmason, J. A. and Sveinsson, J. R., 2005.
Classification of hyperspectral data from urban areas based on ex
tended morphological profiles. IEEE Transactions on Geoscience
and Remote Sensing 42, pp. 480^491.
Brunzell, H. and Eriksson, J., 2000. Feature reduction for clas
sification of multidimensional data. Pattern Recognition 33,
pp.1741-1748.
Duarte-Carvajalino, J., Castillo, P. and Velez Reyes, M., 2007.
Comparative study of semi-implicit schemes for nonlinear dif
fusion in hyperspectral imagery. IEEE Transactions on Image
Processing 16, pp. 1303-1314.
Goetz, A. F. H., Vane, G., Solomon, J. E. and Rock, B. N., 1985.
Imaging spectrometry for earth remote sensing. Science 228,
pp. 1147-1153.
Gomila, C. and Meyer, F., 1999. Levelings in vector spaces.
In: IEEE International Conference on Image Processing, Vol. 2,
Kluwer Academic, pp. 929-933.
Jackway, P. T. and Deriche, M., 1996. Scale-space properties of
multiscale morphological dilation-erosion. IEEE Transactions on
Pattern Analysis and Machine Intelligence 18(1), pp. 38-51.
Karantzalos, K. and Argialas, D., 2006. Improving edge detec
tion and watershed segmentation with anisotropic diffusion and
morphological levelings. International Journal of Remote Sens
ing 27, pp. 5427-5434.
Karantzalos, K., Argialas, D. and Paragios, N., 2007. Compar
ing morphological levelings constrained by different markers. In:
ISMM, G.Banon, et al. (eds), Mathematical Morphology and its
Applications to Signal and Image Processing, pp. 113-124.
Landgrebe, D. A., 2003. Signal theory methods in multispectral
remote sensing. Hoboken: John Wiley and Sons.
Lennon, M., Mercier, G. and Hubert-Moy, L., 2002. Classifica
tion of hyperspectral images with nonlinear filtering and support
vector machines. In: IEEE International Geoscience and Remote
Sensing Symposium, Vol. 3, pp. 1670-1672.
Lindeberg, T., 1994. Scale-Space Theory in Computer Vision.
Kluwer Academic Publishers, Dordrecht.
Maathuis, B. and van Genderen, J., 2004. A review of satellite
and airborne sensors for remote sensing based detection of mine
fields and landmines. International Journal of Remote Sensing
25(23), pp. 5201-5245.
Martin-Herrero, J., 2007. Anisotropic diffusion in the hyper
cube. IEEE Transactions on Geoscience and Remote Sensing 45,
pp.1386-1398.
Meyer, F., 1998. From connected operators to levelings. In:
Mathematical Morphology and Its Applications to Image and
Signal Processing, (H. Heijmans and J. Roerdink, Eds.), Kluwer
Academic, pp. 191-198.
Meyer, F., 2004. Levelings, image simplification filters for seg
mentation. International Journal of Mathematical Imaging and
Vision 20, pp. 59-72.
Meyer, F. and Maragos, P., 2000. Nonlinear scale-space repre
sentation with morphological levelings. Journal of Visual Com
munication and Image Representation 11, PP- 245-265.
Paragios, N., Chen, Y. and Faugeras, O., 2005. Handbook of
Mathematical Models of Computer Vision. Springer.
Plaza, A., Benediktsson, J. A., Boardman, J., Brazile, J., Bruz-
zone, L., Camps-valls, G., Chanussot, J., Fauvel, M., Gamba, P.,
Gualtieri, A., Marconcini, M., Tilton, J. and Trianni, G., 2008.
Recent advances in techniques for hyperspectral image process
ing. Remote Sensing of Environment,(to appear).
Plaza, A., Martinez, P, Plaza, J. and Perez, R., 2005. Dimension
ality reduction and classification of hyperspectral image data us
ing sequences of extended morphological transformations. IEEE
Transactions on Geoscience and Remote Sensing 43, pp. 466-
479.
Schmidt, K. and Skidmore, A., 2004. Smoothing vegetation spec
tra with wavelets. International Journal of Remote Sensing 25(6),
pp.1167-1184.
Tschumperle, D. and Deriche, R., 2005. Vector-valued image
regularization with pdes: A common framework for different ap
plications. IEEE Transactions on Pattern Analysis and Machine
Intelligence 27(4), pp. 506-517.
Vaiphasa, C., 2006. Consideration of smoothing techniques
for hyperspectral remote sensing. International Journal of Pho
togrammetry and Remote Sensing 60, pp. 91-99.
van der Meer, F., 2006. The effectiveness of spectral similarity
measures for the analysis of hyperspectral imagery. International
Journal of Applied Earth Observation and Geoinformation 8(1),
pp. 3-17.
Wang, Z., Bovik, A., Sheikh, H. and Simoncelli, E., 2004. Image
quality assessment: From error visibility to structural similarity.
IEEE Transactions on Image Processing 13, pp. 600-612.
Webb, A., 2002. Statistical Pattern Recognition. John Wiley and
Sons Ltd., UK.
Wilkinson, G., 2003. Are remotely sensed image classification
techniques improving? Results of a long term trend analysis. In:
IEEE Workshop on Advances in Techniques for Analysis of Re
motely Sensed Data, Greenbelt, Maryland, pp. 27-28.