Super-resolution imaging in remote sensing

Abstract

A new effective image super resolution (SR) algorithm which is a hybrid of multiple frame Variational Bayesian (VB) reconstruction and single frame Dictionary Learning (DL) reconstruction method is developed to reconstruct a high resolution (HR) satellite image in this article. Firstly, by employing a variational Bayesian analysis, the unknown high resolution image, the acquisition process, the motion parameters and the unknown model parameters are built together in a single mathematical model with a Bayesian formula, and then the distributions of all unknowns are jointly estimated. Without any parameter adjustment, an HR image is adaptively reconstructed from multiple frame low resolution (LR) images. Secondly, by taking the above HR image as input, a higher resolution image can be rebuilt utilizing the statistical correlation between the HR and LR images which is obtained via the DL method. The VB method effectively uses non-redundant information between LR images to recover HR satellite images. Benefit from the dictionary training of magnanimity image, the DL algorithm is able to provide more high-frequency image details, which means this hybrid of VB and DL method combines the above advantages. The experiments show that this proposed algorithm can effectively increase the image resolution of remote sensing images by 0.5times at least comparing with low resolution image.