Massively parallel image restoration with spatially variant point-spread-function

In this paper we describe a newly developed code, iMPaIR, which performs iterative image deconvolation in parallel. The basic algorithm used is the Richardson- Lucy Maximum-Likelihood iterative procedure. A spatially-invariant point spread function or a spatially-variant point spread function may be used in the deconvolution. The basic Richardson-Lucy olgorithm is described as well as details of the parallel implemention. Applications and results in the areas of astrophysical imaging and medical x-ray imaging are briefly discussed. lit the medical field such restoration algorithms are impractical on single processors-computation time should measured ill seconds rather than hours. WC show that for this type of application a small number of processors should be able to analyse a full X-Ray image (similar to 3730x3062 pixels) in less than a minute.