Contribution to the Joint Workshop of ISPRS WG I/1, I/3 and IV/4
SENSORS AND MAPPING FROM SPACE, Hannover, September 29-October 2, 1997
Boris Prinz, Rafael Wiemker, Hartwig Spitzer
II. Institut für Experimentalphysik
Universität Hamburg
Germany
Mail: B. Prinz / KOGS, Vogt-Kölln-Str. 30, 22527 Hamburg, FRG
WWW-Page: http://kogs-www.informatik.uni-hamburg.de/projects/Censis.html
E-mail: {prinz,wiemker}@informatik.uni-hamburg.de, hartwig.spitzer@desy.de
For the years 1997 to 2000 it is expected that a number of new satellites will be launched into orbit by private companies which are specified to deliver panchromatic imagery of the earth surface with a spatial resolution as fine as 1 m. In contrast to the panchromatic band, the spectrally resolved bands will have a four times coarser ground resolution. Therefore, image fusion algorithms will certainly be employed in order to produce `sharpened' color imagery.
The new satellites have the potential of stimulating and expanding the remote sensing market for image products at a resolution around one meter. In order to prepare for this era we have examined image fusion algorithms using already available airborne imagery. This paper tests fusion algorithms on imagery which was simulated using multispectral images of an airborne scanner (DAEDALUS ATM) with an average resolution of 1 m.
The main advantage of a simulation of satellite images is the possibility to immediately check the deviation between the original 1 m imagery and the fused (1 m + 4 m) multispectral image. Thus, various fusion algorithms can be tested with regard to their accuracy concerning e.g. land cover classification or change detection.
The spectral accuracy of the fused imagery depends strongly on the spatial resolution and scene content. Therefore, accuracy assessments from e.g. SPOT + LANDSAT TM - fusion can certainly not simply be extrapolated down to 1 m imagery. We find that the spectral accuracy of the simulated fused imagery indeed varies with different fusion algorithms. Even though the spectral accuracy of the fused imagery turns out to be limited, we find consistent results of land cover classifications.