De-noising the galaxies in the Hubble XDF with EMPCA

Abstract

We present a method to model optical images of galaxies using expectation maximization principal components analysis. The method returns de-noised images, it relies on the data alone and does not assume any pre-established model or fitting formula. It preserves the statistical properties of the sample, minimizing possible biases. The precision of the reconstructions appears to be suited for photometric, morphological and weak-lensing analysis, as well as the realization of mock astronomical images. As a case study, we discuss the latter because weak gravitational lensing is entering a new phase in which systematics are becoming the major source of uncertainty. Accurate simulations are necessary to perform a reliable calibration of the ellipticity measurements on which the final bias depends. The same is true for strong gravitational lensing where complex morphologies are required to simulate the substructures of strongly magnified galaxies. At this end, the clean images produced with this method can be deconvolved with other means. As a test case for a deep sample, we process 7038 galaxies observed with the Advanced Camera for Surveys Wide Field Channel (ACS/WFC) stacked images of the Hubble eXtreme Deep Field and measure the accuracy of the reconstructions in terms of their moments of brightness, which turn out to be comparable to what can be achieved with weak-lensing algorithms.