Suppressing Random Noise in Mock Galaxy Images through Filtering Methods

Abstract Monte Carlo radiation transfer simulations are widely used to generate mock galaxy images based on hydrodynamical simulation outputs, which is important for connecting physics considered in hydrodynamical simulations to galaxy observables. These images unavoidably contain random noise because practical Monte Carlo calculations can only perform finite size statistical sampling. The straightforward way of suppressing this kind of noise is to use large number of photon packages (≥10 9 ) in radiation transfer calculations, but this leads to heavy computational costs, which limit the potential of applying these simulations onto large theoretical galaxy samples. In this work we investigate another way, namely suppressing this noise through various image filtering methods, including mean, median, Gaussian and bilateral filtering, and filtering based on discrete Fourier transformation and discrete Haar wavelet transformation, with both amplitude and frequency thresholds. We first estimate the noise levels of images from simulations of low number of photon packages (10 7 ), and then observe how they change with the applications of various filtering methods. Several methods can obviously reduce the image noise levels, but not sufficient to replicate the results of simulations run with 10 9 photon packages. An even better filtering method is needed, and it can be obtained probably through adopting a spatial kernel or wavelet more sophisticated than what considered in this work.