Enhanced MRF based Super Resolution Method for Remote Sensing Images

Abstract

In this paper, a learning based enhanced Markov Random Field (MRF) based super resolution reconstruction (SRR) method for remote sensing image with embedded Image Euclidean distance (IMED) is proposed. A robust and transformation invariant similarity metric IMED is integrated for modelling compatibility functions (CF) and finding the similarity between image patches. Unlike traditional Euclidean distance, IMED takes into consideration the spatial relationships of pixels as well as the smallest deformation and therefore provides reasonable result. Further, an iterative belief propagation (BP) algorithm is used to find the optimal candidate patches and therefore high resolution (HR) patches. The experimental results demonstrate that the proposed method outperforms some of the state-of-the-art methods.