Image Processing Reference
In-Depth Information
Chapter 10
Results and Discussions
10.1 Introduction
This chapter provides the consolidated results of various techniques for visualization-
oriented fusion of hyperspectral images presented in Chaps. 3 , 5 - 7 of this monograph.
These fusion techniques together span a wide variety of image processing method-
ologies. We began with a signal processing approach, and derived the
-mattes for
fusion using a bilateral filter. The second technique dealt with fusion as an estimation
problem, and provided a solution using the Bayesian framework. We also explored
the concept of matte-less fusion in Chap. 6 where there was no explicit calculation
of fusion weights. Lastly, we posed fusion as an optimization problem where some
of the desired characteristics of the fused output image had driven the process of
fusion. Combining the results from all these solutions enables us to compare and
analyze their performances. We also consider some of the other recently developed
techniques of hyperspectral image fusion for comparison along with the techniques
discussed in this monograph. The visualization technique related to the selection
of three specific bands chosen according to a certain set of criteria is quite popular
for rendering the contents of the hyperspectral scene. Though such techniques do
not involve any kind of fusion, and have been shown to be inconsistent, these tech-
niques are computationally simpler and easy to operate. We present the results of
the three band selection technique described in [49] along with the other results for
the performance assessment. Two very specific techniques for displaying the hyper-
spectral image have been proposed in [78, 79]. These techniques generate the fused
image from a linear combination of the spectral bands. In the first technique, the
fusion weights have been calculated from a set of piecewise linear functions which
are based on certain optimization criteria for display on the standard color space
(sRGB), and the perceptual color space (CIELAB). The second technique developed
by the same authors defines the fusion weights from the CIE 1964 tristimulus func-
tions by stretching them across the wavelength axis in order to cover the entire range
of the hyperspectral image. We provide the results of fusion of hyperspectral images
using these two techniques for comparisons. The quantitative performance indices
Search WWH ::

Custom Search