Image Processing Reference
In-Depth Information
has already been carried out, there is no reason why the candidate band should then
be discarded.
8.3 Experimental Results
In this section, we discuss some experimental results for the output-based band selec-
tion scheme over the two datasets- the urban data, and the moffett
2
data which the
readers are familiar with. The performance of this scheme is dependent on the fusion
technique employed. Although, our motivation of developing the output-based band
selection comes from the output-based fusion scheme presented in the last chapter,
purposes. There are two reasons for the choice of fusion technique. First, we have
provided the experimental results for the input-based band selection scheme using
the performance of the band selection scheme using various measures over the out-
put images fused using the bilateral filtering-based technique. Thus, providing the
resultant images using the same fusion technique over the bands selected using the
output-based scheme will facilitate the readers to compare the performances of both
the band selection schemes. Second, the output-based band selection requires gener-
ation of the so-called intermediate fusion output for every additional band selected.
The bilateral filtering-based fusion technique is a non-iterative, faster process as
opposed to the optimization-based fusion which is iterative in nature. Thus, the
bilateral filtering-based fusion technique turns out to be a quick, yet reliable option
for the analysis and demonstration purposes.
Figure
8.1
a shows the result of fusion of the urban data by the Hyperion using the
bilateral filtering-based technique. The resultant image is, however, obtained from
the combining of only 27 bands out of a total of nearly 170 useful bands in the
original data. One may compare this figure to Fig.
8.1
b representing fusion of the
entire urban data cube. This figure is essentially the same as the corresponding fused
image in Chap.
4
as the fused technique and the set of bands being fused are exactly
the same. It may be observed that Fig.
8.1
a brings out most of the features as does
the other image, and does not introduce any visible artifacts in the result. Thus, the
resultant images fused over the subsets of the hyperspectral data selected using the
output-based scheme are comparable to the resultant images produced from fusion
of the entire dataset using the same fusion technique in terms of visual quality.
As we have already introduced two parameters for the performance evaluation of
fused images
viz.
, the entropy, and the average gradient, we shall continue to use the
same in the present chapter. More details on evaluation of fusion techniques will be
The entropy represents the information content in the image. Figure
8.2
ashows
the entropies of the fused images for different values of the threshold parameter
κ
.
As
increases, the number of bands selected for fusion reduces, and thus, one may
expect smaller values of the entropy as fewer input bands are contributing towards
κ
