Super Resolution and Image Enhancement

Abstract

Super resolution (SR) is a technique to estimate a high-resolution (HR) image from one or several low-resolution (LR) images. SR can be broadly classified into two families of methods: (i) The classical multi-frame super-resolution, and (ii) Example-based or learning-based super-resolution, which is also known as single-frame super-resolution. In the classical multi-frame SR, the HR image is reconstructed by combining subpixel-aligned multi-images (LR images). Since the reconstruction of HR image from LR images is often an ill-posed problem, we need to include some a prior knowledge or additional assumptions for reconstruction. If sparsity is used as a prior knowledge, the SR can be considered as one of compressed sensing techniques. In the learning-based single-frame SR, the HR image is estimated by learning correspondence between low- and high-resolution images from a database. In this paper, we introduce a multi-frame SR and a single-frame SR for medical image enhancement. Both of them use the sparsity as a prior knowledge.