Image Processing Reference
comprising a bandwidth of 735.190mm, has been shown in Fig. 3.5 a. The subsequent
resultant image as shown in Fig. 3.5 b represents fusion of the next one-third block
of the moffett 2 dataset corresponding to the bands from 79-152 of the original data.
The total bandwidth encompassed by these bands, and thus, by the corresponding
fused image in Fig. 3.5 b is 706.875nm. The remaining subset of the hyperspectral
bands forms the input for the third fused image at this stage as shown in Fig. 3.5 c.
It may be noted that these bands have been initially fused into six first-level fused
images, which are then combined to obtain the second-level fused image. This image
essentially combines the remaining 72 bands with a total bandwidth of 705.140nm.
Fig. 3.5 d depicts the final fused image obtained through fusion of three resultant
images from the pre-final stage, i.e., the second-level fusion stage in this case. This
image has the total bandwidth of 2146.780nm which is the sum of the individual
bandwidths of the constituent images at any given stage.
This chapter explains a solution to fusion of hyperspectral image for the purpose of
visualization. This solution uses an edge preserving bilateral filter for defining the
fusionweight based on the locally dominant features. The discussed solution provides
flexibility in the process of fusion by modifying the bilateral filter parameters, and the
tuning parameter C to match the needs of the observer. As the bilateral filter is non-
iterative and efficient implementation schemes are available, this solution is fast and
computationally simple. Furthermore, this technique does not require construction
of multi-resolution pyramids. As it requires the fusion weights to be computed for
each and every pixel of the hyperspectral data cube, the fusion process exploits all
available information at all locations in all bands, unlike the techniques where the
entire band receives the same weight.
This chapter also discusses a hierarchical implementation scheme for fusion so as
to accommodate any increase in the number of bands without degradation in fusion
performance. Additionally, this scheme enables visualization and analysis of fusion
of bands up to any given spectral bandwidth.