Image Processing Reference

In-Depth Information

Table 9.2
Performance measures for various techniques for visualization of the geological data

( c

2011 Elsevier)

Fusion technique

Relative bias

Fusion factor

Fusion symmetry

b

F F

F S

Bilateral filtering technique

0.21

1.59

0.19

Three band selection

0.10

1.33

0.44

Piecewise linear function

0.24

1.71

0.38

Color matching function

0.23

1.75

0.35

MRA technique

0.18

1.36

0.16

For a comparison of the numerical values, we provide the results of the participa-

tory performance measures for the final images
F
, using all these studied techniques

for theMoffett data Table
9.1
, and the Geological data Table
9.2
. The multi-resolution

analysis (MRA)-based technique turns out to be superior to most of the other tech-

niques in terms of relative bias
b
and fusion symmetry FS. However, a very low value

of the fusion factor FF indicates a smaller amount of mutual information between

the final image and the constituent bands. Therefore, although the participation from

the image bands is quite uniform, the quantitative gain from the bands in terms of

information is less as indicated by low values of the fusion factor.

The bilateral filtering-based technique performs well in terms of low values of

the relative bias and fusion symmetry, indicating that the final image has a lower

derivation from the constituent images in terms of intensities, as well as information

content. High values of fusion factor FF indicate a significant amount of mutual

information between the constituent bands and the fused image.

The techniques using piecewise linear functions and color matching functions

provide results with a high variance yielding a good contrast in the final results

(will be discussed in the next chapter). The objective assessment shows that both

techniques offer a high value of fusion factor (FF), but values of the fusion symmetry

FSare also large, which is not desirable. This indicates an uneven participation from

input hyperspectral bands. Thus, not all the bands contribute uniformly toward the

fusion process. Additionally, the values of the fusion symmetry parameter for these

two techniques were found to vary over the test datasets as well. The technique of

piecewise linear function and the color matching function assign a constant fusion

weight to each of the constituent bands. In case of the piecewise linear function

technique, these fixed weights are derived from the optimization of certain criterion

in the sRGB space, while the technique of color matching function assigns these

constant weights to the particular band by modeling certain perceptual properties of

the human eye. In both cases, however, the calculation of weights is fixed and does

not consider the local variation in data contents at all. Quite naturally, the subset of

bands receiving higher weights has a higher contribution towards the fused image.

Therefore, features from this subset of bands find a good representation in the fused

image. On the other hand, features in rest of the bands are likely to contribute very

little as the corresponding bands have smaller fusion weights. Therefore, one can