LiveNet: Improving features generalization for face liveness detection using convolution neural networks
作者：
Highlights：
• An efficient training strategy for training deep CNN with limited data is proposed.
• The technique is based on continuous data randomization in the training dataset.
• The training time required for CNN reduces by 18.39%.
• The HTER is reduced by 8.28% on CASIA Database and 14.41% on Reply Attack Database.
• Detail experimental analysis are provided on both intra and cross database tests.
摘要
•An efficient training strategy for training deep CNN with limited data is proposed.•The technique is based on continuous data randomization in the training dataset.•The training time required for CNN reduces by 18.39%.•The HTER is reduced by 8.28% on CASIA Database and 14.41% on Reply Attack Database.•Detail experimental analysis are provided on both intra and cross database tests.
论文关键词：Convolution neural networks,Face anti-spoofing,Face liveness detection,Face-biometric,VGG-11,Bootstrapping,EER,HTER
论文评审过程：Received 8 January 2018, Revised 24 March 2018, Accepted 5 May 2018, Available online 8 May 2018, Version of Record 14 May 2018.
论文官网地址：https://doi.org/10.1016/j.eswa.2018.05.004
