首先对计算机人脸自动识别技术的研究背景及发展历程做了简单回顾,然后对人脸正面像的识别方法,按照识别特征的不同进行了分类综述,主要介绍了特征脸(Eigenface)方法、基于小波特征的弹性匹配(Elastic Matching)的方法、形状和灰度模型分离的可变形模型(Flexible Model)以及传统的部件建模等分析方法.通过对各种识别方法的分析与比较,总结了影响人脸识别技术实用化的几个因素,并提出了研究和开发成功的人脸识别技术所需要考虑的几个重要方面,进而展望了人脸识别技术今后的发展方向.

In this paper, Research background of automatic face recognition and its relation to human vision system are briefly reviewed. Then current face recognition technologies are roughly introduced and classified according to different recognition features. Four main algorithms are analyzed and compared. The first is eigenface, which is extraction of global features using the PCA. In this approach, a set of faces is represented using a small number of global eigen vectors, which encode the major variations in the input set. The second is flexible model, which separate shape and gray parameter. The third is wavelet-based elastic graph matching, in which memorized faces are represented by regular graphs, whose vertices are labeled by a multi resolution description in terms of localized spatial frequencies. Spatial relationships within the object are labeled by geometrical distance vectors. The last method is traditional analytical techniques. Based on the analysis and comparison, key factors in face recognition technologies are concluded and distilled as suggestion to future research.