用于可靠身份认证的唇语识别

Lip-reading Algorithm for Reliable Authentication

本文介绍了一种基于人工神经网络的唇语识别系统,可以用来提高身份识别系统的安全性。比较了基于音频和视频的视频序列分割法,并自制了一些唇语样本。为了提高对文中陌生人集团成员和自制样本的唇语识别率,采用了迁移学习的方法。结果表明,通过迁移学习后得到的网络对陌生人集团样本和自制样本有较好的唇语识别率,同时也检验了文中视频序列分割的效果。

This paper introduces a lip-reading system based on artificial neural network which can be used to improve the security of identification system.The ways based on audio and video to segment video sequence are compared and some lip-reading samples are made.To boost the lip-reading recognition rate to the members of stranger group and self-made samples,transfer learning is taken.The results show that the networks gotten after transfer learning have better lip-reading recognition rates to the samples of stranger group and self-made samples,and the effect of video sequence segmentation in this paper is examined at the same time.