Mars Image Super-Resolution Based on Generative Adversarial Network

Abstract

High-resolution (HR) Mars images have great significance for studying the landform features of Mars and analyzing the climate on Mars. Nowadays, the mainstream image super-resolution methods are based on deep learning or CNNs, which are better than traditional methods. However, these deep learning based methods obtain low-resolution(LR) images usually by using an ideal down-sampling method (e.g. bicubic interpolation). There are two limitations in the existing SR methods: 1) The paired LR-HR data by using such methods has a satisfactory results when tested on an ideal datasets. But, these methods always fail in real Mars image super-resolution, since real Mars images rarely obey an ideal down-sampling rule. 2) The LR images obtained by bicubic interpolation have no noise while real Mars images usually have noise, which leads to the super-resolved images are not realistic in texture details. To solve the above-mentioned problems, in this article, we propose a novel two-step framework. Specifically, to address limitation 1), we focus on designing a new degradation framework by estimating blur-kernels. To address limitation 2), a Generative Adversarial Network (GAN) is trained to generate noise distribution. Extensive experiments on the Mars32k dataset demonstrate the effectiveness of the proposed method, and we achieve better qualitative and quantitative results compared to other SOTA methods.INDEX TERMS Generative Adversarial Network, Kernel estimation , Mars image super-resolutionn, Noise model.

Multi-Aperture Joint-Encoding Fourier Ptychography for a Distributed System

“…The former are used for extracting non-redundant, sub-pixel information from multi-frame images [17]; however, they also face the problems of computational speed and the fixed orbit limitation of the camera load. Deep learning-based methods, on the other hand, are data-driven approaches used to improve the spatial resolution of images while maintaining or enhancing their details and quality; however, such methods also suffer from issues such as synthetic textures, the realism of the results, the limitations of the training data, and the generalization and interpretability of the models [18,19].…”