A public parallel method for astronomical image restoration ⋆

Any recorded image is blurred whenever the instrument used to obtain it has a finite resolving power. For example, the image of a point source seen through a telescope has an angular size which is inversely proportional to the diameter of the primary mirror of the telescope. If the instrument is ground-based, the image is additionally degraded by the turbulent motions in the earth’s atmosphere. Image restoration is a useful technique relevant in many fields: medicine and astronomy among others. Basic methods applied to image restoration include filtering, speckle deconvolution and blind deconvolution. Several sequential software approaches are available to tackle the problem. Most of them are provided as commercial software packages. Among others, we can mention the implementations on IDL [1], Pixon [2] or Memsys 5 (an implementation of a maximum entropy method). In the case of astronomy, much effort is presently devoted to the improvement of the spatial resolution of images, either via the introduction of new observing techniques (e.g. interferometry or adaptive optics [3]) or via a subsequent numerical processing of the image (deconvolution). It is, in fact, of major interest to combine both methods to reach an even better resolution. The scientific motivation of such effort is to further exploit the present available data and to reduce the uncertainty on some of the conclusions obtained from the images. In this work we present work in progress in the design of a parallel image restoration system based on a novel algorithm. The main contribution in our approach resides in the minimization of an analytical upper bound to the algorithmic complexity of the image. The details in ground telescope images deconvolved with our algorithm are comparable than those from the Hubble Space Telescope, which is considered as the de facto best imaging telescope up to date. With the new algorithm, our aim is to target astronomical images recorded at ground based telescopes and satellites. Nevertheless, the method is general enough to be applied to any kind of spatial invariant blurred image iff the statistics of the image is known. In our target scenario (astronomical images), the complexity of the problem grows with the number of pixels at the image by the