Applications to Real Measured Data - Introduction to Imaging from Scattered Fields - page 139

Re( x 1 )

(a) (b) (c)

0

2
4
6
8
10
12
14
16

0.06 0.06

1.5

1.5
1.4
1.3
1.2
1.1

0.04
0.02

0.04
0.02

1.4

0

0 1.3

-0.02
-0.04
-0.06

-0.02
-0.04
-0.06

1.2

1.1

0 5 10 15 -0.05 0 0.05

-0.05 0 0.05

(d) (e) (f )

100

4.5
4
3.5
3
2.5
2
1.5
1

50

0

-50

-100

-100 -50 0 0 100

0 1 0 0.6

x (mm)

y

y

(g) (h) (i)

1.7
1.6
1.5
1.4
1.3
1.2
1.1
1

100

50

x

x

0

-50

-100

-3.3

0.0

0.75 0.0

Re{τ( x , y )} Im{τ( x , y )}

-100 -50 0

mm

50 100

Figure 9.25 Comparison of reconstruction algorithms on FoamMetExt: (a) CSI method, (b) MGM method, (c)
MGM method with adaptive multiscale, (d) DTA/CSI methods, (e) Bayesian inversion method (real), (f) Bayesian
inversion method (imaginary), (g) IMSA (Real) method, (h) IMSA (imaginary) method, and (i) cepstral method.

and a quantitative image. Figures 9.25e and 9.25f show the real and imaginary
part of the reconstructions using the Bayesian inversion method (Feron et al.,
2005). Figures 9.25 g and 9.25 h show real and imaginary part of metal object
using IMSA (Donelli et al., 2005). Both the reconstructions from Feron et al.
(2005) and Donelli et al. (2005) seem to lack quantitative recovery. In addition,
they seem to fail to recover all object features in a single image. The image
from the cepstral method seems to show good shape and permittivity estima-
tion, although it still seems to suffer from incorrect background permittivity,
which is likely due to limited data availability.

9.4 FInAl ReMARkS And SuMMARy
One might ask why the reconstructions in the previous using various meth-
ods applied to measured data are relatively poor. This is to some extent the
case even for weakly scattering objects. If we look more carefully at the data
provided by AFRL or the Institut Fresnel, these were made available to assist
with the development and improvement of inverse scattering algorithms, but,