高安全性人脸识别系统中的唇语识别算法研究

发布时间：2018-03-11 19:17

本文选题：高安全性　切入点：人脸识别　出处：《计算机应用研究》2017年04期 　论文类型：期刊论文

【摘要】：针对目前人脸识别系统面临的图片和视频攻击,构建了一种将人脸识别与口令密码相结合,并采用唇语识别技术进行活体检测的高安全性身份认证系统。首先由于汉语唇语数据的缺少,建立了CNLIP1和CNLIP2两个较大的汉语唇语数据库;其次,为了保留唇语的时序性,采用堆叠卷积独立子空间分析(ISA)深度神经网络模型来实现唇动时序特征的提取;最后提出使用迁移学习算法来训练特定人唇语识别模型。实验证明,唇动时序特征能更好地表征出数字串唇语,迁移学习训练的特定人唇语模型能够满足活体检测的需要,所构建的高安全性人脸识别系统具有较好的防攻击效果。
[Abstract]:Aiming at the image and video attacks facing face recognition system, a new method is proposed to combine face recognition with password cipher. Firstly, because of the lack of Chinese lip language data, two large Chinese lip-language databases, CNLIP1 and CNLIP2, are established. Secondly, in order to preserve the temporal characteristics of lip language, a high-security identity authentication system is developed for lip-language recognition in vivo. An ISA-depth neural network model is used to extract the temporal features of lip motion. Finally, a migration learning algorithm is proposed to train a specific human lip recognition model. The lip movement temporal feature can better represent the digital lip string. The specific human lip model trained by transfer learning can meet the needs of living body detection. The high security face recognition system has a better anti-attack effect.
【作者单位】：中国科学院重庆绿色智能技术研究院智能多媒体技术研究中心;
【基金】：中国科学院战略性先导科技专项基金资助项目(XDA06040103) 国家自然科学基金资助项目(61472386);国家自然科学基金青年基金资助项目(61502444)
【分类号】：TP391.41

【相似文献】