Few-shot prototype alignment regularization network for document image layout segementation

作者：

Highlights：

• A novel method for the document image layout analysis problem is proposed.

• It makes better use of the information of the support set by metric learning.

• It learns classification prototype within an embedding space.

• Prototype alignment regularization term between support and query sets is developed.

摘要

•A novel method for the document image layout analysis problem is proposed.•It makes better use of the information of the support set by metric learning.•It learns classification prototype within an embedding space.•Prototype alignment regularization term between support and query sets is developed.

论文关键词：Meta-learning,Few-shot learning,Metric learning,Semantic segmentation

论文评审过程：Received 19 September 2020, Revised 10 January 2021, Accepted 5 February 2021, Available online 11 February 2021, Version of Record 21 February 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.107882