Image Processing Reference
In-Depth Information
Table 10.3
Performance measures for various techniques for visualization of the moffett
3
data
Fusion technique
Variance
Entropy
Avg gradient
Relative
Fusion
Fusion
bias
b
σ
2
H
g
¯
factor
FF
symmetry
FS
Bilateral filtering technique
374.29
5.60
4.62
0.31
1.75
0.21
Bayesian technique
451.84
5.17
4.82
0.32
1.93
0.94
Variational technique
385.35
5.63
3.53
0.18
1.95
0.26
Optimization technique
527.12
5.71
4.47
0.40
2.03
0.98
Three band selection
363.64
5.24
3.30
0.36
1.29
0.49
Piecewise linear function
282.05
5.55
3.18
0.45
1.47
0.47
Color matching function
157.50
5.19
2.59
0.44
1.24
0.28
Table 10.4
Performance measures for various techniques for visualization of the lunar
2
data
Fusion technique
Variance
Entropy
Avg gradient
Relative
Fusion
Fusion
bias
b
2
σ
H
g
¯
factor
FF
symmetry
FS
Bilateral filtering technique
642.93
6.32
4.59
0.24
1.65
0.08
Bayesian technique
630.91
6.47
4.43
0.19
1.42
0.04
Variational technique
611.25
6.41
4.22
0.14
2.12
0.36
Optimization technique
669.91
6.40
5.02
0.23
2.14
0.15
Three band selection
658.49
6.47
4.35
0.21
1.83
0.11
Piecewise linear function
642.35
6.45
4.61
0.16
1.70
0.04
Color matching function
612.49
6.56
4.47
0.16
2.15
0.06
formed by the specifically selected three bands as shown in Fig.
10.3
e is capable of
capturing most of the data contents. The bilateral filtering-based solution particularly
enhances the weaker features in the data, and therefore, provides somewhat higher
amount of visual information than the other techniques. These minor differences in
the quality of the fusion results are observable from the quantitative performance
measures provided in Table
10.4
. All the techniques generate fused images with sim-
ilar values of variance and entropies. The fusion factor FF and fusion symmetry FS
of these techniques indicate comparable participation of bands for all the techniques.
The result of variational technique possesses the smallest value of the relative bias
by virtue of the iterative correction in the mean of the fused image, which in turn
measures the deviation of the output intensity from that of input bands. Higher value
of the fusion factor for this technique also indicates a large amount of information
transfer from the input bands towards the fusion output.
The optimization-based technique yields the highest value of the average gradient
than any other technique, which results in a better perceived sharpness in Fig.
10.3
d.
This technique processes the hyperspectral data for obtaining an image with high
contrast, irrespective of the input characteristics. Therefore, we are able to obtain a
high value of the variance in the result despite lack of contrast within the input bands.
As we may note, this advantage comes at the cost of a higher value of fusion symmetry.
Having discussed the results of processing the AVIRIS data, we consider the
remaining three hyperspectral datasets provided by the Hyperion imaging sensor.
