Image Processing Reference
In-Depth Information
explained in Chap.
9
bring uniformity to such comparisons. This chapter presents a
comparative study using the same set of performance measures.
Each of the fusion solutions requires a certain set of parameters which control the
quality of the fused image. We have used the same parameter values for the implemen-
tation of each of the fusion solutions described in respective chapters of this mono-
graph. For other techniques, we have used the same parameters as described in the
corresponding references cited therein. While it is possible to obtain either grayscale
or RGB results of the fusion using any of the fusion techniques from the mono-
graph, most of the performance measures have been defined for the grayscale images.
Thus, we discuss the quantitative assessment of the resultant grayscale images of all
the techniques considered in this chapter. The RGB images, however, facilitate an
easy and better visual comparison. Therefore we also illustrate the color results of
the fusion techniques discussed here. The readers are requested to go through the
e-book version of the monograph for an RGB pseudo-color representation of the
fused images. Also, the appearance of the color images varies with the settings of
display and printing devices. The readers are requested to take a note of this.
10.2 Description of Hyperspectral Datasets
For any of the fusion solutions discussed in this topic, we do not assume any prior
knowledge about the scene or the imaging system. The original hyperspectral image
contains nearly 200
+
bands. However, some of these bands have a near zero response
as the molecular absorption bands of water and carbon dioxide block the transmission
of radiation in certain wavelength bands [160]. In addition to those, data in some
bands are highly corrupted which appear like noise. After removal of these bands, the
number of available bands reduces to around 175-180. Additionally, we normalize
the hyperspectral data, i.e., the dynamic range of the pixel has been scaled to unity
through an appropriate scaling of the whole dataset.
In this chapter, we provide extensive experimental results for 6 datasets from
two different hyperspectral devices. The first hyperspectral sensor is the Airborne
Visible/Infrared Imaging Spectrometer (AVIRIS) operated by the National Aero-
nautics and Space Administration/Jet Propulsion Laboratory (NASA/JPL).
1
The
datasets provided by the AVIRIS contain 224 bands where the size of each band
is
pixels. The first two datasets by the AVIRIS depict some regions of
the Moffett Field, CA. We refer to these data as the moffett
2
, and the moffett
3
data,
respectively. The moffett
2
data have already been discussed in the last chapter for
illustrations of the consistency of fusion techniques. Another dataset provided by the
AVIRIS captures the area of the Lunar lake, NV. These data which depict underwater
features have been referred to as the lunar
2
in this chapter.
(
614
×
512
)
1
AVIRIS data available from JPL/NASA:
http://aviris.jpl.nasa.gov
.
