International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
114
(a) (b)
Fig. 5. Overlay of vectorised forest masks to aerial reference image, © BEV 1992. (a) Forest mask derived from
Landsat TM image, (b) Forest mask derived from AIF image, © Joanneum Research 1999.
between the fusion results and the “true” image could be
performed. The evaluation proved the spectral stability of the
algorithm and the higher correlation between the “true” and the
fused image compared to the degraded image.
The third case study showed the application of AIF to
multisensor image data acquired by Landsat TM and SPOT
PAN. The objective was the derivation of a forest mask using a
threshold technique. While the thresholding of the Landsat TM
bands provided a reliable mask, the shape of this mask could be
significantly improved by first applying the AIF to the
multispectral bands and the panchromatic image.
The conclusions drawn from the application in the test sites are
manifold. The AIF is considered useful for areas that are
dominated by objects larger than the low resolution pixel size.
Taking the average pixel size of current multispectral sensors
this prerequisite is only true for certain types of image objects,
such as agricultural fields. However, with the increasing spatial
resolution of the new sensor generation, this limitation will
become less and less important. On the contrary, the application
of AIF might help to reduce the enormous complexity of very
high resolution images. In cases of small objects that do not
benefit from the AIF, we suggest to use the fusion result for first
level classification, i.e. classification of meta-classes that
provide image masks for subsequent detailed classification. The
advantage of this approach lies in the sharper delineation of the
meta-objects, thus leading to less misclassifications caused by
mixed pixels.
A final suggestion refers to the combination of AIF and
substitution techniques for visualisation. AIF would first
sharpen object edges and thus eliminate the blocky pixel
structure of the low resolution image. Then the local texture
could be added by applying a substitution technique onto the
AIF processed image, thus leading to a sharpened high
resolution multispectral image product. However, this product
is not considered an appropriate input for numerical
classification but offers an improved basis for visual
interpretation of the image.
ACKNOWLEDGEMENTS
Part of this work has been supported by the European
Commission, Contract No. ENV4-CT96-0359. The author
thanks W. Pillmann (OBIG) for providing the ATM data of
Vienna. The contribution of Ursula Schmitt (Joanneum
Research) to the forest case study is greatfully acknowledged.
REFERENCES
Benediktsson J.A. and Swain P.H. 1992. Consensus theoretic
classification methods. IEEE Trans. Syst., Man, Cybem., 22(4),
pp. 688-704.
Carlson G.R. and Patel B. (1997): A new area dawns for geo
spatial imagery. GIS World, 10(3), pp. 36-40.
Carper W.J., Lillesand T.M. and Kiefer R.W., 1990. The use of
intensity-hue-saturation transformation for merging SPOT
panchromatic and multispectral image data. Photogrammetric
Eng. Remote Sensing, 56(4), pp. 459-467.
Chavez P.S., Sides S.C. and Anderson J.A., 1991. Comparison
of three different methods to merge multiresolution and
multispectral data: Landsat TM and SPOT panchromatic.
Photogrammetric Eng. Remote Sensing, 57(3), pp. 295-303.
Janssen L.L., Jaarsma M.N. and van der Linden E.T., 1990.
Integrating topographic data with remote sensing for land-cover
classification. Photogrammetric Eng. Remote Sensing, 56(11),
pp. 1503-1506.
Garguet-Duport B., Girel J., Chassery J.-M., Pautou G., 1996.
The use of multiresolution analysis and wavelet transform for
merging SPOT panchromatic and multispectral image data.
Photogrammetric Eng. Remote Sensing, 62(9), pp. 1057-1066.
Harris J.R., Murray R. and Hirose T., 1990. IHS Transform for
the integration of radar imagery with other remotely sensed data.
Photogrammetric Eng. Remote Sensing, 56(12), pp. 1631-1641.
