Remote Sensing Image Inpainting with Generative Adversarial Networks

Andrey Kuznetsov, Mikhail Gashnikov

Abstract: We investigate generative adversarial neural networks (GAN) for remote sensing image inpainting. We are considering a generative neural network with a contour predictor. We use this neural network to inpainting of the natural remote sensing images obtained by “Canopus”, “Meteor”, “AIST”, “Resurs” aircrafts, as well as Google Earth images. As a basis for comparison, we use an exemplar-based algorithm. We experimentally prove the effectiveness of the generative neural network with the contour predictor for remote sensing image inpainting, in particular for generation forgery Earth remote sensing data.