Enhancing convolutional neural networks for face recognition with occlusion maps and batch triplet loss

作者:

Highlights:

• Identify which regions of a face a CNN relies on to achieve a high recognition rate.

• Occlude these face regions during training to increase robustness to face occlusions.

• Enhance triplet loss by minimising the standard deviation of the score distributions.

• Improved results using the proposed methods on both the AR and the LFW datasets.

摘要

•Identify which regions of a face a CNN relies on to achieve a high recognition rate.•Occlude these face regions during training to increase robustness to face occlusions.•Enhance triplet loss by minimising the standard deviation of the score distributions.•Improved results using the proposed methods on both the AR and the LFW datasets.

论文关键词:Face recognition,Convolutional neural networks,Facial occlusions,Distance metric learning

论文评审过程:Received 13 September 2017, Revised 29 March 2018, Accepted 12 September 2018, Available online 22 September 2018, Version of Record 4 October 2018.

论文官网地址:https://doi.org/10.1016/j.imavis.2018.09.011