Data driven model based least squares image reconstruction for radio astronomy

Image reconstruction problems in radio astronomy and other fields like biomedical imaging are often ill-posed and some form of regularization is required. This imposes user specified constraints to the reconstruction process that may produce an undesirable bias to the solution. We propose a data driven model based least squares reconstruction method based on the Karhunen-Loève transform. We show that this constraint stems from intrinsic physical properties of the measurement process and demonstrate the improvement of the method over unregularized least squares reconstruction using actual data from the Low Frequency Array (LOFAR), a phased array radio telescope in the Netherlands.