Facet-Based Regularization for Scalable Radio-Interferometric Imaging

Current and future radio telescopes deal with large volumes of data and are expected to generate high resolution gigapixel-size images. The imaging problem in radio interferometry is highly ill-posed and the choice of the prior model of the sky is of utmost importance to guarantee a reliable reconstruction. Traditionally, one or more regularization terms (e.g. sparsity and positivity) are applied for the complete image. However, radio sky images can often contain individual source clusters in a large empty background. Exploiting this observation, we propose a smarter regularization term. More precisely, we propose to divide radio images into source occupancy regions (facets) and apply relevant regularizing assumptions for each facet. Leveraging a stochastic primal dual algorithm, we show the potential merits of applying facet-based regularization on the radio-interferometric images which results in both computation time and memory requirement savings.