Deep autoencoders for attribute preserving face de-identification

作者：

Highlights：

• An autoencoder-based face de-identification method is proposed

• Both supervised (attribute information) and unsupervised (no labels) methods proposed

• Three metrics are proposed for experimental evaluation of the proposed method

• The proposed method can achieve near perfect de-identification with realistic results

• Very lightweight architecture, suitable for deployment on embedded devices

摘要

•An autoencoder-based face de-identification method is proposed•Both supervised (attribute information) and unsupervised (no labels) methods proposed•Three metrics are proposed for experimental evaluation of the proposed method•The proposed method can achieve near perfect de-identification with realistic results•Very lightweight architecture, suitable for deployment on embedded devices

论文关键词：Face de-identification,Attribute preservation,Deep learning,Autoencoders

论文评审过程：Received 13 May 2019, Revised 30 September 2019, Accepted 9 November 2019, Available online 15 November 2019, Version of Record 20 November 2019.

论文官网地址：https://doi.org/10.1016/j.image.2019.115699