A Robust Image Super-solution Algorithm Using Deep Convolution Networks

Abstract

The aim of single image super-resolution (SISR) is to find a non-linear mapping from the low to high resolution training image samples and reconstruct a high-resolution (HR) image from a low-resolution (LR) image using the prior knowledge. As a powerful tool in solving high-level computer vision problems, deep learning can also be used to solve the low-level vision problem, such as SISR. By exploring the architecture of convolutional neural network (CNN)，we put forward a robust SISR algorithm based on CNN. Comparison to the state-of-art SISR methods, the proposed SISR algorithm obtains the best SR performance.