Deep Post-Processing for Sparse Image Deconvolution
Published in Signal Processing with Adaptive Sparse Structured Representations (SPARS) workshop, 2019
Abstract
Variational-based methods are the state-of-the-art in sparse image deconvolution. Yet, this class of methods might not scale to large dimensions of interest in current high resolution imaging applications. To overcome this limitation, we propose to solve the sparse deconvolution problem through a two-step approach consisting in first solving (approximately and fast) an optimization problem followed by a neural network for “Deep Post Processing”(DPP). We illustrate our method in radio astronomy, where algorithms scalability is paramount due to the extreme data dimensions. First results suggest that DPP is able to achieve similar quality to state-of-the-art methods in a fraction of the time.