人脸自动识别是模式识别和图像处理等学科的一大研究热点, 在身份鉴别、信用卡识别、护照核对以及监控系统等方面有着广泛的应用.提出一种基于概率主成分分析方法(PPCA)的人脸识别,该方法与传统的主成分分析(PCA)相比,克服了简单的"丢弃"其他非主成分因子,在PPCA中将"丢弃"因子作为噪声成分进行估计,同时PPCA方法是一种基于概率模型的方法,因此很容易延伸为混合模型,对于PPCA概率模型参数,提出利用EM算法对其进行估计.用两个不同的数据集(姿势表情变化集和光照变化集),将PPCA人脸识别算法和传统的PCA算法进行比较,基于PPCA的人脸识别算法中的"丢弃"方差的收敛速度快于传统的PCA算法.实验结果表明,无论是姿势表情变化集,还是光照变化集,PPCA算法的识别率都优于传统的PCA识别算法.

Automatic face recognition was an active research area in the last decade. With the increased importance of security and organization, identification and authentication methods were developed into a key technology in various areas such as entrance control in building. Face recognition method based on probabilistic principle component analysis (PPCA) was proposed. However, a notable feature of traditional PCA was the absence of an associated probabilistic model for the observed data. A probabilistic formulation of PCA from a Gaussian latent variable model was obtained, which was closely related to statistical factor analysis. The parameters of PPCA could be determined using EM algorithm. In experiments, the proposed methods have been successfully evaluated using two different datasets. The experimental results show that the face recognition method based on PPCA is superior to the method based on the traditional PCA.