SRNHARB: A deep light-weight image super resolution network using hybrid activation residual blocks

作者：

Highlights：

• A novel residual block that extracts and processes both the positive and negative-valued features is proposed for the task of image super resolution.

• The proposed residual block is designed in a light-weight manner by employing group-wise convolution operations.

• A two-stage strategy using two different loss functions is employed for training the proposed network in order to enhance the super resolution performance.

摘要

•A novel residual block that extracts and processes both the positive and negative-valued features is proposed for the task of image super resolution.•The proposed residual block is designed in a light-weight manner by employing group-wise convolution operations.•A two-stage strategy using two different loss functions is employed for training the proposed network in order to enhance the super resolution performance.

论文关键词：Image super resolution,Deep learning,Residual learning

论文评审过程：Received 1 May 2021, Revised 17 August 2021, Accepted 16 September 2021, Available online 25 September 2021, Version of Record 29 September 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116509