Introduction - Hyperspectral Image Fusion - page 4

grayscale image, the airborne or the satellite sensor captures incoming radiation only
at a given range of wavelength, and discards the rest of the electromagnetic (EM)
spectrum. The discarded part of the EM spectrum contains variations in the amount
of reflected and/or emitted radiation at those wavelengths.
If this incoming radiation is collected over a sufficiently narrow spectral band, one
obtains an average radiance of the underlying material integrated over that particular
spectral band. When the bandwidth of each of the spectral bands is sufficiently small,
of the order of a few nanometers, one can practically obtain an almost continuous
spectral response of the material as a function of wavelength. This response function
is unique for any given material, and hence, in principle, can be used for identification
purposes. This array of spectral observations is referred to as the spectral signature
of the material. The spectral signature is one of the most important characteristics of
the material for several reasons:

￿ Being unique, it provides unambiguous information towards the identification of

the materials in the scene.

￿ A typical image captures the materials present on the outer surface of the earth. The

spectral signature gets constituted from the radiance pattern of the materials that
could be present beneath the surface of the earth, and thus not directly visible. This
property makes the spectral signature particularly useful in the task of identifying
the minerals beneath the surface of the earth.

￿ When there exists a mixture of two or more underlying materials within a sin-

gle pixel, the captured spectral signature is comprised of the weighted sum of
the spectral signatures of the individual materials. Using some spectral unmixing
techniques, one can identify various materials present along with their proportions.
When the spectral response of the scene is captured by a single wideband sen-
sor, the magnitude of the image pixel represents the integrated spectral response of
the scene material. In this case, one obtains a single image of the scene—that is an
integrated response of the scene over the bandwidth of the imaging sensor. Natu-
rally, such images lack in their ability to discriminate among various materials. This
shortcoming of the single sensor imaging gave rise to the idea of spectral imaging.
The idea of spectral imaging is based on the use of spectral variations in the
electromagnetic field. Multispectral (MS) image captures the spectral information
of the scene in the form of around 4-6 bands covering the visible, reflective infrared
(IR), and thermal IR regions of the spectrum. Each of the bands represents radiation
information acquired over a fairly wide-band in the wavelength spectrum. Thus, one
can think of the multispectral data as the sampled version of the spectral response
of the pixel. For most multispectral imaging systems, however, these bands are few,
wide and separated; and therefore, the sampling of the spectral response is not dense.
The multispectral (MS) images have several advantages over a single sensor image.
The MS image captures spectral information from the bands that are beyond the
visible range. This set of bands captures the reflectance response of the scene over the
particular spectral bands, and thus, each band provides complementary information
about the scene with respect to the other bands. With multiple observations of the
scene element, the identification and classification capabilities of the MS data are