Rafael E. Carrillo, Jason D. McEwen, Yves Wiaux
We extend previously proposed radio-interferometric imaging approaches based on convex optimization to handle continuous visibilities and large-scale optimization problems. We propose a general algorithmic framework based on the simultaneous-direction method of multipliers to solve sparse imaging problems. The algorithm offers a parallel implementation structure, thus providing a significant gain in terms of speed and scalability to very high dimensions. We implement various state-of-the-art sparsity regularization priors, including our recent average sparsity approach SARA, in a new imaging software dubbed PURIFY. We evaluate through realistic simulations the performance of the software in terms of reconstruction quality and computational speed. Simulation results confirm both the superiority of SARA for continuous Fourier measurements and the fact that the new algorithmic structure offers a promising path to handle large-scale problems. Code is available at https://github.com/basp-group/purify
View original:
http://arxiv.org/abs/1307.4370
No comments:
Post a Comment