The Multi-scale Fast Network For Image Super-resolution

Single image super resolution (SISR), in recent years, has made great progress. By studying the existing image super-resolution algorithm, we find that the existing algorithm has a disadvantage, that is, the network is very deep and the training is too difficult. In order to solve this problem, we have improved based on EDSR and proposed a multi-scale fast network model for image super-resolution (MFSR). The network consists of multi-level feature extraction (MLFE), multi-scale feature fusion (MSFF), and upsampling reconstruction. The network can fully extract input image features with reducing network depth. MFSR improves the difficulty of training existing networks. Experiments show that our network has achieved good performance.