Single Remote Sensing Image Super-Resolution and Denoising via Sparse Representation

Abstract

This paper presents a new approach to generate a high-resolution (HR) remote sensing image from a single low-resolution (LR) input while denoising simultaneously, based on sparse signal representation. Recent research on patch-based sparse representation suggests that the high resolution patch has the same sparse representation as the corresponding low resolution patch. Inspired by this observation, we jointly train two dictionaries for the low resolution and the high resolution image patches and enforce the similarity of sparse representations between them. Thus using Batch Orthogonal Matching Pursuit (Batch-OMP), we seek a sparse representation for each patch of the low-resolution input which can be applied with the high resolution dictionary to generate a high resolution patch. We first adopt sparse representation in the area of remote sensing image super-resolution and denoising, with state-of-the-art performance, equivalent and sometimes surpassing other SR methods recently published.