Video Super-Resolution via Residual Learning

Abstract

Convolutional neural networks have been widely applied in many low level vision tasks. In this paper, we propose a video super-resolution (SR) method named enhanced video SR network with residual blocks (EVSR). The proposed EVSR fully exploits spatio-temporal information and can implicitly capture motion relations between consecutive frames. Therefore, unlike conventional methods to video SR, EVSR does not require an explicit motion compensation process. In addition, residual learning framework exhibits excellence in convergence rate and performance improvement. Based on this, residual blocks and long skip-connection with dimension adjustment layer are proposed to predict high-frequency details. Extensive experiments validate the superiority of our approach over state-of-the-art algorithms.

Keywords

• Video super-resolution
• convolutional neural network
• implicit motion compensation
• residual block
• skip-connection