Constrained and unconstrained algorithms of the multisensor multiresolution
technique (MMT) are discussed. They can be applied to unmix low-resolution
images using the information about their pixel composition from co-registe
red high-resolution images. This makes it possible to fuse the low- and hig
h-resolution images for a synergetic interpretation. The constrained unmixi
ng preserves all the available radiometric information of the low-resolutio
n image. On the other hand, the unconstrained unmixing may be preferable in
case of noisy data. An analysis of the MMT sensitivity to sensor errors sh
owed that the strongest requirement is the accuracy of geometric co-registr
ation of the data; the co-registration errors should not exceed 0.1-0.2 of
the low-resolution pixel size. Applications of the constrained and unconstr
ained algorithms are illustrated on examples of unmixing and fusion of the
multiresolution reflective and thermal bands of a real TM/LANDSAT image as
well as of a simulated image of the future ASTER/EOS-AM1 sensor.