MuRNet: A deep recursive network for super resolution of bicubically interpolated images
作者：
Highlights：
• A new recursive block for the problem of single image super resolution that generates a very rich set of features resulting from different spatial ranges and different resolution levels is proposed in order to provide a state-of-the-art performance.
• Through a careful choice of the number of filters, the total number of parameters of the recursive block, and hence, that of the network is kept low.
• The recursion process provides the network with a sufficient effective depth and yet the number of operations is kept low in view of the small number of parameters used by the recursive block.
摘要
•A new recursive block for the problem of single image super resolution that generates a very rich set of features resulting from different spatial ranges and different resolution levels is proposed in order to provide a state-of-the-art performance.•Through a careful choice of the number of filters, the total number of parameters of the recursive block, and hence, that of the network is kept low.•The recursion process provides the network with a sufficient effective depth and yet the number of operations is kept low in view of the small number of parameters used by the recursive block.
论文关键词：Image super resolution,Recursive convolutional neural networks,Deep learning
论文评审过程：Received 7 June 2020, Revised 25 September 2020, Accepted 4 March 2021, Available online 13 March 2021, Version of Record 18 March 2021.
论文官网地址：https://doi.org/10.1016/j.image.2021.116228
