Back to Volume
Paper: Parallelism for HST Image Restoration -- Some Experiences and Options
Volume: 25, Astronomical Data Analysis Software and Systems I
Page: 215
Authors: Adorf, H.-M.; Oldfield, M. J.
Abstract: We report on recent experience gained in a project set up to explore the power of parallel computing for the restoration of Hubble Space Telescope images. Using our vector-processor implementation of the standard SIPSF Richardson-Lucy algorithm, we achieve a full 40-iteration restoration of a 512x512 image frame in less than 90 seconds. With this considerable computing power we carry out Monte Carlo simulations which allow the quantification of the probable statistical errors in a restoration. We compare the relative merits of our approach, which utilizes fine-grained parallelism, with several alternatives including a low-cost one transforming a workstation cluster into a coarse-grained ``virtual supercomputer''.
Back to Volume