Image Processing Reference
In-Depth Information
obtain very good fusion resultswith such techniqueswhere the subset of bands getting
a higher weightage coincides with the subset of hyperspectral data containing high
amounts of information. However, in other cases, these techniques may unnecessarily
highlight a subset of bands does not contain enough useful features. In contrast to
this, the bilateral filtering-based technique and the MRA-based technique evaluate
the local neighborhood of each and every pixel in the data for the calculation of fusion
weights. Hence, the later techniques prove to be more consistent with the quality of
the result produced.
For visual illustration of the
consistency
of a technique, we provide some of the
incrementally fused images of the moffett
2
dataset using the bilateral filtering-based
technique. The result of fusion of first 50 bands of the moffett
2
dataset can be seen
in Fig.
9.5
a. The resultant image from fusion of 100 bands of the same data has been
provided in Fig.
9.5
b. A comparison between these two images gives an idea about
the contribution of subsequent bands. A fusion of first 50 bands provides the outline
of the rough shape of the objects present in the data, while subtle details and finer
boundaries appear clear in Fig.
9.5
b where 100 bands have been combined. This
effect is more clear in the top left and right corners of the image which contain a
(a)
(b)
(c)
(d)
Fig. 9.5
Results of bilateral filtering-based incremental visualization of the moffett
2
data.
a
,
b
and
c
show the results of fusion of 50, 100, and 150 bands, respectively.
d
shows the final fused image
( c
2011 Elsevier, Ref: [92])
