Light field reconstruction using hierarchical features fusion

作者：

Highlights：

• U-Net can architecture to extract hierarchical features.

• SAS can extract spatial and angular features for the 4D light field.

• U-Net and SAS are combined into U-SAS-Net to take advantages of the two.

• U-SAS-Net achieves 0.62 db better and 10x faster than state-of-the-art.

• A large patch size is proved to be beneficial for learning the light field.

摘要

•U-Net can architecture to extract hierarchical features.•SAS can extract spatial and angular features for the 4D light field.•U-Net and SAS are combined into U-SAS-Net to take advantages of the two.•U-SAS-Net achieves 0.62 db better and 10x faster than state-of-the-art.•A large patch size is proved to be beneficial for learning the light field.

论文关键词：Light field,Deep learning,Neural network,Image processing

论文评审过程：Received 15 September 2019, Revised 6 December 2019, Accepted 16 March 2020, Available online 18 March 2020, Version of Record 10 April 2020.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.113394