This paper deals with the reconstruction of the boundary and the permittivi
ty of homogeneous objects from synthetic data. The method is based on a rig
orous integral method and a conjugate gradient algorithm requiring the solu
tion of only two direct problems at each step. To model some noise resultin
g from the inhomogeneity of the background, this object is surrounded by a
set of point-like scatterers, arbitrarily distributed between the two paral
lel lines of measurement. Since convergence of the algorithm strongly depen
ds on the initial guess, a method to detect a target in a random medium has
been adapted. Statistical numerical results for reconstruction in a crossw
ell configuration are also presented.