Single image super resolution based on learning features to constrain back projection

Abstract

Image super-resolution (SR) is an active research point due to its added value for many image processing applications. The classical SR aims to obtain a high resolution (HR) image using multiple low resolution (LR) images. Recently many research works are directed towards obtaining such HR image from a single LR image which is known as single image SR restoration.This paper presents a fast single-image SR approach based on learning the functions that can transfer LR patch into HR features. Then, these features are used to reconstruct the HR image through a process called constrained back-projection. The experimental results show that the proposed approach is capable of providing a high quality super-resolution images.