Recursive residual atrous spatial pyramid pooling network for single image deraining

作者：

Highlights：

• We propose a recursive residual atrous spatial pyramid pooling network for single image deraining.

• ResASPP module is introduced to utilize the multi-scale features of rainy image, which can enlarge the receptive field.

• Recursive learning is used to strengthen the model capability by multi-stage deraining processing from coarse to fine while maintain the model parameters.

• Extensive experiments on both real-world and synthetic datasets demonstrate that our approach significantly outperforms state-of-the-art in image deraining task.

摘要

•We propose a recursive residual atrous spatial pyramid pooling network for single image deraining.•ResASPP module is introduced to utilize the multi-scale features of rainy image, which can enlarge the receptive field.•Recursive learning is used to strengthen the model capability by multi-stage deraining processing from coarse to fine while maintain the model parameters.•Extensive experiments on both real-world and synthetic datasets demonstrate that our approach significantly outperforms state-of-the-art in image deraining task.

论文关键词：Deraining,Recursive neural network,Multi-scale,Residual atrous spatial pyramid pooling

论文评审过程：Received 13 August 2020, Revised 14 July 2021, Accepted 10 August 2021, Available online 18 August 2021, Version of Record 26 August 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116430