A Basic Study on Multimodal Biometric Person Authentication : Signature, Face and Voice

用于验证个人身份的算法大致可以分为以下四类：静态、动态、生物识别和基于物理/知识的。例如，指纹、虹膜、视网膜、DNA、面部和血管都是静态的和生物特征的。被归类为生物识别和动态识别的算法包括嘴唇动作、身体动作和在线签名。使用密码的方案是静态的和基于知识的，而使用IC卡、磁卡和钥匙的方法是物理的。本研究项目调查了以下生物识别方案作为迈向多模式身份验证的第一步：1.在线签名我们提出了几种不同的算法。最新的一种是用马尔可夫链蒙特卡罗(MCMC)的贝叶斯方法。同时实现了0.93%的FRR和0.73%的FAR。人脸从自然图像数据中分割出人脸的子图像后，使用Gabor小波特征结合MCMC对9个特征点进行估计。已经观察到比以前的方案有了很大的改进。已经尝试了几种算法。这三种方法的整体融合仍处于起步阶段，将是未来项目的主题。

The algorithms used to verify personal identity can roughly be classified into the following four categories : static, dynamic, biometric, and physical/knowledge-based. For instance, fingerprints, iris, retina, DNA, face and blood vessels are static and biometric. Algorithms classified as biometric and dynamic include lip movements, body movements and on-line signatures. Schemes that use passwords are static and knowledge-based, whereas methods using IC cards, magnetic cards, and key are physical.This research project investigated the following biometric schemes as the first step toward multimodal authentication :1. online signatureWe have proposed several different algorithms. The most recent one uses Bayesian method with MCMC (Markov Chain Monte Carlo). FRR of 0.93% and FAR of 0.73% were simultaneously achieved.2. faceAfter segmenting a face subimage from a natural image data, nine feature points were esitiamted using Gabor wevelette features together with MCMC. A great improvements over the previous sheme has been observed.2. voiceSeveral algorithms have been tried.Overall fusion of the three methods are still in its infancy and will be a topic of future project.