摘 要随着科技时代的到来,个人身份的虚拟化以及人们对于个人身份安全的需求越来越高。怎么确认登陆系统的人员身份,且不给登陆的用户带来太过繁琐的流程,是现代系统需要考虑的核心问题。面部识别技术由于具有唯一性且方便快捷,使得其在金融、通行、网络安全等领域,得到了飞速的技术进步。但是,当这种识别技术得到普及的时候,一些图谋不轨的不法分子也逐渐开始寻找和利用面部识别的系统漏洞,来破解登陆系统,这使得信息认证受到了挑战。而系统的开发这只能通过在登陆系统的时候,对用户提出随机动作,例如点头和眨眼等,以此来确认登陆系统的不是图像和机器,但是这样的作法,不能很好的解决系统的缺陷。因而人们开始将研究重心进行了转移,在研究如何检测真人的方法中,加入唇语识别技术。每个人的嘴唇形状和纹理都是不一样的,哪怕有人的嘴唇形状相似,但是说话方式是也肯定不一样。由于唇语识别在身份认证上有着得天独厚的优势,所以我们接下来讨论唇语识别的原理和使用到的技术。本文主要先将唇部区域进行分割,然后用 AdaBoost训练器训练,最后通过卷积神经网络的的 Tensorflow 框架进行处理。关键词:唇语识别;卷积神经网络;身份认证AbstractBecause of the development of the technological age, the virtualization of personal identities and people's demand for personal identity security are increasing. How to confirm the identity of the person who logs in the system, and does not bring too complicated processes to the logged-in user, is the core issue that modern systems need to consider. Facial recognition technology is unique, convenient and fast, which has made rapid technological progress in the fields of finance, traffic and network security. However, when this kind of recognition technology became popular, some criminals with bad intentions gradually began to find and use the system vulnerability of facial recognition to crack the login system, which made the information authentication challenged. The development of the system can only be done by presenting random actions to the user when logging int...