Image Reconstruction Based on Group Sparsity Representation in Radio Interferometry

Astronomical images exhibit strong nonlocal self-similarity. Extended radio structures, galactic shells, symmetric jets, and background emissions produce abundant similar patches across the field of view. Different from traditional fixed sparse bases applied to the entire astronomical image, the GSRI method employs groups containing nonlocal image patches with similar structures as the basic unit to sparsely represent the image.

In addition, a low-complexity adaptive dictionary learning scheme is designed for each group. Furthermore, we adopt the split Bregman method to efficiently solve the L0 minimization problem.Results. We conduct several experiments on the Very Large Array and DAocheng Radio Telescope.

The GSRI approach is compared with several state-of-the-art imaging algorithms. Experimental results demonstrate its superior performance in both reconstruction accuracy and computational speed.