Image Processing Reference
In-Depth Information
(e)
(f)
(g)
Fig. 10.2
(continued)
The relative bias
b
and the fusion symmetry (FS) for the band selection and the
optimization-based techniques are quite high for the same reasons explained in the
case of the moffett
2
data. The bilateral filtering-based technique and the variational
technique do not deviate from the mean of the input hyperspectral image as opposed
to the other two techniques where this deviation is explicitly imposed for an efficient
visualization. The variational technique provides the least deviation in terms of the
relative bias
b
as the technique has been designed to maintain the overall saturation
in the fused image close to that of the input hyperspectral bands.
Let us now consider the lunar
2
dataset from the AVIRIS. This dataset is different
from the previous datasets of the Moffett Field in terms of scene contents. This dataset
depicts some underwater features of the Lunar lake. The results of fusion using the
techniques discussed in this monograph are shown in Fig.
10.3
. We observe that
almost all these results are visually similar. The lunar data do not contain much fea-
tures as opposed to the first two datasets. Additionally, most of these features are
present over a substantially large number of spectral bands. Therefore, the composite
