obtained at the same degree of convergence without any regularization, with a Tikhonov (TK) regularization, and with our EP regularization scheme. The reconstruction without regularization shows a blurred profile, with a coarse shape description. The use of a TK regularization smoothes the profile, and the edges are not preserved. The new regularization scheme improves the perfonnance of the CG algorithm: the edges are clearly preserved, while the homogeneous areas are smoothed.We proposed, then, two other results, obtained from two mystery objects, named IPS005 and IPS007. The scattered fields were collected from 36 viewing angles, 9, ~[0",350"], with a sample spacing of lo", over the observation sector 8, SOs <8,+170",withasamplespacingof AOs =lo". Theonly one piece of information given with these data sets was the radius of the minimum circumscribing circle. We show, in Figure 2 and Figure 3, the results obtained by using the CG algorithm, without any regularization (no a priori information used), and also the corresponding original profiles revealed at the 1996 Symposium. The domain, L, is discretized into 29 x 29 square cells of 1 mm', for target IPS005, and of 0.5 mm2, for target IPS007. The main problem was to find a satisfying calibration factor for each mystery target. After several blind tests, we were able to propose two suitable reconstructions, using valid but non-optimum calibration parameters. These reconstructions gave quite good spatial resolution of the t w~ mystery objects.
ConclusionThis paper presents some evidence of the effectiveness of adding the EP regularization to the CG method, in reconstructing the shape and permittivity profile of dielectric objects. Without any a priori information, our CG algorithm is also still efficient, and succeeded in reconstructing two mystery targets. As the two targets are now known, we hope we can greatly enhance the reconstruction quality by choosing better values for the calibration factors, and by applying our EP regularization to these data.
References1. P. Lobel, R. Kleinman, Ch. Pichot, L.