The information that can be extracted from an image of a galaxy is fundamentally limited by the resolution and noise in the data. Schawinski et al. have applied a machine learning method to galaxy images, which is trained by comparing artificially degraded images with the originals. The algorithm is then used to recover features from previously unseen degraded images, which it performs more successfully than traditional deconvolution techniques. The method requires assuming that the target galaxies look similar to those in the training set, and individual details can be lost or misidentified, but it should be useful for studying statistical properties of galaxies in large surveys.
Mon. Not. R. Astron. Soc.467, L110 (2017).