Generation of Starry Sky Images by Generative Adversarial Networks

Abstract

In recent years, the rapid development of machine learning has attracted attention to the generation of creative works by artificial intelligence. In particular, in the field of images, Generative Adversarial Networks has made it possible to edit and generate high quality images in various domains. However, in the case of starry sky photographs, while the sequence of stars is one of the important factors, there is not enough data set on the positional relationship of each star to generate the exact sequence of stars. Therefore, the editing and generation of starry sky photographs by artificial intelligence is a difficult task. In this study, we propose a method to generate images similar to actual starry sky photographs by using Generative Adversarial Networks and the previously proposed method of creating star maps as a preliminary step, with the final goal of automatic editing of starry sky photographs by artificial intelligence. It has been cofirmed that the generated images are comparable to actual starry sky photographs.