Emanuele Salerno

... Anna Tonazzini, Luigi Bedini, Ercan E. Kuruoglu and Emanuele Salerno ... 5, pp. 3617-3620, 1997. [9] E. Kuruoglu, L. Bedini, MT Paratore, E. Salerno, A. Tonazz-ini, ”Source separation in astrophysical maps using indepen-dent factor analysis”, Neural Networks, to appear, 2003. ...

This paper addresses the problem of microwave tomographic imaging from far-field monostatic measurements. In the first-order Born approximation, a simple Fourier relationship exists between the measured data and dielectric contrast of the object under test. For a lossy object, the contrast function depends on frequency, and this prevents monostatic data from being used directly since multifrequency probing radiation must be used to get sufficient information. To overcome this difficulty, we propose a data preprocessing strategy that eliminates frequency from the function to be reconstructed. The Fourier data obtained through preprocessing could be immediately inverted to give a bandpass estimate of the unknown function. Despite the drawbacks inherent in this approach, in small-scale tomographic applications, the space region occupied by the object under test, as well as the sign of the unknown function, are normally known. This extra information allows us to apply a projected Landweber algorithm to uniquely reconstruct the unknown, thus avoiding the above-mentioned drawbacks. We outline the preprocessing and reconstruction techniques adopted, and show some preliminary results from experimental and simulated data

We present a very simple monostatic setup for coherent multifrequency microwave measurements, and an optimization procedure to reconstruct high-resolution permittivity profiles of layered objects from complex reflection coefficients. This system is capable of precisely locating internal inhomogeneities in dielectric bodies, and can be applied to on-site diagnosis of architectural components. While limiting the imaging possibilities to 1D permittivity profiles, the monostatic geometry has an important advantage over multistatic tomographic systems, since these are normally confined to laboratories, and on-site applications are difficult to devise. The sensor is a transmitting-receiving microwave antenna, and the complex reflection coefficients are measured at a number of discrete frequencies over the system passband by using a general-purpose vector network analyzer. A dedicated instrument could also be designed, thus realizing an unexpensive, easy-to-handle system. The profile reconstruction algorithm is based on the optimization of an objective functional that includes a data-fit term and a regularization term. The first consists in the norm of the complex vector difference between the measured data and the data computed by a forward solver from the current estimate of the profile function. The regularization term enforces a piecewise smooth model for the solution, based on two 1D interacting Markov random fields: the intensity field, which models the continuous permittivity values, and the binary line field, which accounts for the possible presence of discontinuities in the profile. The data-fit and the regularization terms are balanced through a tunable regularization coefficient. By virtue of this prior model, the final result is robust against noise, and overcomes the usual limitations in spatial resolution induced by the wavelengths of the probing radiations. Indeed, the accuracy in the location of the discontinuities is only limited by the system noise and the discretization grid used by the forward solver. The algorithm we chose to optimize the objective is based on the particle swarm paradigm. Each feasible solution is coded as a location in a multidimensional space, explored by a number of "particles" each moving with a certain velocity, which is partly random and partly induced by the experience of both the particle itself and the "swarm" of all the other particles. In our case, the search is complicated by the mixed continuous-binary nature of our unknowns, but the swarm intelligence approach maintains the advantage of its intrinsic parallelism. The experimental results we obtained from both simulated and real measurements show that, for typical permittivity values and radiation wavelengths, the spatial resolution is highly improved by the line process. From real measurements in the range 1.7-2.6 GHz, we accurately reconstructed the permittivity values of our test phantom and located the discontinuities within the limits imposed by our discretization grid (with 1.5 mm cell thickness). At present, the applicability of our reconstruction method is still limited by the forward solver, which is based on a cascaded transmission-line model that assumes normal and plane-wave incidence. We are developing a new solver based on a closed-form Green's function in multilayered media, which should enable us to model appropriately both the microwave sensor and the illumination geometry, thus improving the accuracy of the computed reflection coefficients in the objective functional.

... of Information Engineering, University ofPisa, Via Caruso, 56122 Pisa. e-mail: a.monorchio(g. manara)@iet.unipi.it ... 1697-1708,July 2000. [3] H. Hidalgo, JL Marroquin, E. Gomez-Trevino, "Piecewise Smooth Models for Electromagnetic Inverse Problems", IEEE Trans. Geosci. ...

This paper proposes a semi-blind approach for the separation of sources that are not totally uncorrelated. Up to now, all work on separation of components in astrophysical maps have assumed statistically independent sources. In this work, we attempt to perform a dependent component analysis of astrophysical sources and we propose a semi-blind algorithm which provides separation using second-order statistics only.

ABSTRACT Planck has mapped the intensity and polarization of the sky at microwave frequencies with unprecedented sensitivity. We make use of the Planck 353 GHz I, Q, and U Stokes maps as dust templates, and cross-correlate them with the Planck and WMAP data at 12 frequencies from 23 to 353 GHz, over circular patches with 10 degree radius. The cross-correlation analysis is performed for both intensity and polarization data in a consistent manner. We use a mask that focuses our analysis on the diffuse interstellar medium at intermediate Galactic latitudes. We determine the spectral indices of dust emission in intensity and polarization between 100 and 353 GHz, for each sky-patch. The mean values, $1.63\pm0.03$ for polarization and $1.52\pm0.02$ for intensity, for a mean dust temperature of 18.7 K, are close, but significantly different. We determine the mean spectral energy distribution (SED) of the microwave emission, correlated with the 353 GHz dust templates, by averaging the results of the correlation over all sky-patches. We find that the mean SED increases for decreasing frequencies at $\nu < 60$ GHz, for both intensity and polarization. The rise of the polarization SED towards low frequencies may be accounted for by a synchrotron component correlated with dust, with no need for any polarization of the anomalous microwave emission. We use a spectral model to separate the synchrotron and dust polarization and to characterize the spectral dependence of the dust polarization fraction. The polarization fraction ($p$) of the dust emission decreases by $(34\pm10)$ % from 353 to 70 GHz. The decrease of $p$ could indicate differences in polarization efficiency among components of interstellar dust (e.g., carbon versus silicate grains), or, alternatively, it could be a signature of magnetic dipole emission from ferromagnetic inclusions within interstellar grains.

A second-order statistical technique (FD-CCA) for semi-blind source separation from multiple-sensor data is presented. It works in the Fourier domain and allows us to both learn the unknown mixing operator and estimate the source cross-spectra before applying the proper source separation step. If applied to small sky patches, our algorithm can be used to extract diffuse astrophysical sources from the

Blind Source Separation techniques, based both on Indepen- dent Component Analysis and on second order statistics, are presented and compared for extracting partially hidden texts and textures in doc- ument images. Barely perceivable features may occur, for instance, in ancient documents previously erased and then re-written (palimpsests), or for transparency or seeping of ink from the reverse side, or from

A color decorrelation strategy to improve the human or automatic readability of degraded documents is presented. The particular degradation that is considered here is bleed-through, that is, a pattern that interferes with the text to be read due to seeping of ink from the reverse side of the document. A simplified linear model for this degradation is introduced to permit

Abstract This paper deals with motion segmentation. We use an active contour method to segment the apparent motion computed from two successive frames of an image sequence. In order to obtain accurate segmentation, we combine both intensity and motion ...

In this paper we review a maximum a posteriori (MAP) approach detection method in a Bayesian scheme which incorporates prior information about the source flux distribution, the locations and the number of sources of extragalactic point sources in images of the Cosmic Microwave Background. This new technique allows us to obtain fast solutions and to fix the number of detected

The problem of edge-preserving tomographic reconstruction from Gaussian data is considered. The problem is formulated in a Bayesian framework, where the image is modeled as a pair of Markov Random Fields: a continuous-valued intensity process and a binary line process. The solution, defined as the maximizer of the posterior probability, is obtained using a Generalized Expectation-Maximization (GEM) algorithm in which