Fast and reliable probabilistic face embeddings based on constrained data uncertainty estimation

作者：

Highlights：

• A simplification of the MLS metric is proposed to speed up the matching process.

• A unilateral constraint loss is proposed to improve the stability of recognition.

• A feature fusion method is proposed for video face recognition.

• Experiments show that the proposed method can achieve SOTA performance with a fast matching speed.

摘要

•A simplification of the MLS metric is proposed to speed up the matching process.•A unilateral constraint loss is proposed to improve the stability of recognition.•A feature fusion method is proposed for video face recognition.•Experiments show that the proposed method can achieve SOTA performance with a fast matching speed.

论文关键词：Probabilistic face embeddings,Deep learning,Face recognition,Data uncertainty estimation

论文评审过程：Received 29 June 2021, Revised 30 January 2022, Accepted 8 March 2022, Available online 21 March 2022, Version of Record 29 March 2022.

论文官网地址：https://doi.org/10.1016/j.imavis.2022.104429