1558 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 46, NO. 5, MAY 2008
A Nonlinear Iterative Reconstruction and Analysis
Approach to Shape-Based Approximate
Electromagnetic Tomography
Naren Naik, Jerry Eriksson, Pieter de Groen, and Hichem Sahli
Abstract—A nonlinear Helmholtz-equation-modeled electro-
magnetic tomographic reconstruction problem is solved for the
object boundary and inhomogeneity parameters in a damped
Tikhonov-regularized Gauss–Newton (DTRGN) solution frame-
work. In this paper, the object is represented in a suitable
global basis, whereas the boundary is expressed as the zero level
set of a signed-distance function. For an explicit parameterized
boundary-representation-based reconstruction scheme, analytical
Jacobian and Hessian calculations are made to express the changes
in scattered field values w.r.t. changes in the inhomogeneity pa-
rameters and the control points in a spline representation of the
object boundary, via the use of a level-set representation of the
object. Even though, in this paper, a homogeneous dielectric is
considered and a spline representation has been used to represent
the boundary, the formulation can be used for a general global
basis representation of the inhomogeneity as well as arbitrary
parameterizations of the boundary, and is generalizable to three
dimensions. Reconstruction results are presented for test cases
of landminelike dielectric objects embedded in the ground under
noisy data conditions. To confirm convergence and, at times, to
know which of the obtained iterates are closer to the actual
unknown solution, using a perturbation theory framework, a local
(Hessian-based) convergence analysis is applied to the DTRGN
scheme for the reconstruction, yielding estimates of convergence
rates in the residual and parameter spaces.
Index Terms—Boundary reconstruction, Gauss–Newton
method, local analysis, nonlinear tomography, subsurface
imaging, Tikhonov regularization.
I. I NTRODUCTION
T
HE NEED for nonlinear reconstruction approaches oc-
curs in many branches of electromagnetic tomographic
imaging. Some examples are wave-equation-based diffraction
tomography, electrical impedance tomography, and diffusion
optical tomography, to name just a few. The area of tomo-
Manuscript received May 30, 2006; revised September 30, 2007. This work
was supported in part by the Vrije Universiteit Brussel Research Council
under Concerted Action Project GOA-20 on “Numerical issues in tomographic
shallow subsurface imaging.”
N. Naik is with the Department of Mechanical Engineering and the College
of Engineering, University of Canterbury, Christchurch 8140, New Zealand
(e-mail: naren.naik@canterbury.ac.nz; nnaikt@yahoo.com).
J. Eriksson is with the Department of Computing Science, Umeå University,
901 87 Umeå, Sweden (e-mail: jerry@cs.umu.se).
P. de Groen is with the Department of Mathematics, Vrije Universiteit
Brussel, 1050 Brussels, Belgium (e-mail: pdegroen@vub.ac.be).
H. Sahli is with the Department of Electronics and Informatics (ETRO), Vrije
Universiteit Brussel, 1050 Brussels, Belgium, and also with the Interuniver-
sitair Micro-Elektronica Centrum, B-3001 Leuven, Belgium (e-mail: hsahli@
etro.vub.ac.be).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TGRS.2008.916077
graphic subsurface imaging is of interest in important applica-
tions such as geophysical prospecting and landmine detection,
to name only a couple. Typically, the reconstruction of the
desired parameters is achieved in the following possible ways:
1) by a minimization of an objective functional comprising
the residual of the measured and modeled data when either
the full scattering operator [1]–[3] or an approximate operator
such as a first- or a second-order Born approximation [4], [5] is
considered and 2) via analytical methods, which are based on
approximations of the forward operator [6].
The minimization class of problems, being ill posed in na-
ture, needs additional a priori information (such as support of
the object) to aid the convergence of the solution scheme to
the desired correct solution. The motivations of using boundary
information in a tomographic reconstruction are typically to
lend stability to the iterative reconstruction scheme (by better
demarcation of object support constraints and also by the pos-
sible reduction of the number of parameters characterizing the
object) or to solve an “approximate” inverse scattering problem
wherein the object shape, location, and an approximate (as
against a more exact) estimate of the object’s interior physi-
cal parameter values are reconstructed. The iterative “shape-
based” approximate reconstruction schemes broadly fall into
two categories. The objective functional minimized in the first
class has as unknowns the coefficients in an explicit parametric
representation for the boundary curve(s), whereas, in the latter
class, the unknowns are the values of a set function representing
the image, with the zero level set of that function representing
the boundary. While the first (explicit representation) class of
schemes has the advantage of fewer unknowns, which is useful
in potential 3-D reconstructions, the second (implicit represen-
tation) class is better suited to handle topological changes in
the evolving shape of the boundary. In this paper, we have
formulated a scheme for the solution and convergence analysis
of the nonlinear 2-D approximate reconstruction problem in the
first class of schemes, in a framework for arbitrary parameter-
izations, that is conceptually generalizable to three dimensions
as well. We use a level-set representation for the shapes in
order to calculate the first- and second-order shape and elec-
tromagnetic parameter derivatives of the objective functional in
a parameterized representation and not in the more customary
implicit level-set representation of the boundaries.
Miller et al. [7] solved the approximate inverse scattering
problem using a partially nonlinear Born iterative scheme with
a spline representation of the image boundary and a global basis
representation for the object and background inhomogeneities.
They use a greedy-search-type technique to find the best change
0196-2892/$25.00 © 2008 IEEE