Image Processing Reference

In-Depth Information

up the results from all the fusion techniques together facilitates a better comparative

analysis to the readers both in a subjective and objective terms. Therefore, we shall

present combined results of fusion over several hyperspectral datasets using these

techniques explained in this monograph along with some other recently developed

techniques in a later chapter. However, in this chapter we provide the results of fusion

using the Bayesian technique over a single dataset for a quick illustration.

Consider the urban hyperspectral data used in the previous chapter again. These

data depict some region of Palo Alto, CA that has several urban establishments.

The data contain 242 bands of dimensions

each. We initially remove

the noisy and zero-response bands as done for the previous solution. We process the

entire dataset to obtain a grayscale fused image. First we compute the sensor selectiv-

ity factor

(

512

×

256

)

for every pixel in the data which is the normalized product of two quality

measures. The second step consists of an iterative procedure to generate the fused

image
F
. The fused image so obtained represents the best possible image estimated

under the given constraints for the given hyperspectral data. This resultant image, as

shown in Fig.
5.2
a is a combined response of the scene over an approximate band-

width of 2200 nm. The fusion results are also often provided as color (RGB) images

as they provide a better visual interpretation of the scene. Most of the fusion solu-

tions for hyperspectral data generate the color images using some pseudo-coloring

schemes. In order to illustrate the RGB versions, we first partition the input data into

three subsets. While there could be various strategies for this partitioning, we follow

a simple one. We obtain three groups by sequentially partitioning the data along the

wavelength such that every groups contains nearly a similar number of bands. These

three groups are processed independently using the Bayesian technique to generate

the corresponding three fused images. These images are then assigned to the red,

β

Fig. 5.2
Results of Bayesian

fusion of the urban image from

the Hyperion.
a
Grayscale

fused image, and
b
RGB

fused image (©2013 Elsevier,

Ref. [91]). (Color is viewable

in e-book only.)

(a)

(b)