通过创建虚拟样本的小样本人脸识别统计学习方法

研究了人脸识别方法．在传统的“特征脸”方法基础上，提出了一种基于贝叶斯方法的小样本人脸识别方法，该方法对于经过预处理的标准人脸图像，通过原型脸创建虚拟样本，扩充样本数量，然后用PCA降维并提取人脸图像的特征．对提取的特征用BEM算法学习该类样本的概率密度分布参数，构建贝叶斯混合网络分类器．该方法可以有效地解决统计学习方法中样本数量不足问题，提高小样本人脸识别方法的识别率，同样可以运用于模式识别中其它对象识别．实验表明，该方法能提高小样本人脸识别率，有实际应用价值。

A SMALL SAMPLE FACE RECOGNITION STATISTICAL LEARNING METHOD BASED ON VIRTUAL SAMPLES

The face recognition approach is studied. A small sample face recognition approach based on Bayes is presented. In this method, prototype faces are used to create virtual samples in order to expand the sample number. Then, the PCA method is used to reduce dimensions and extract features. BEM algorithm learns the parameters of condition probability function of class features. A Bayes classifier is constructed to recognize face. The method combines the principal component analysis technique and the Bayes classifier and shows their feasibility on the face recognition problem. It also overcomes the problem of limited samples in pattern recognition application and improves the recognition rate of small samples face recognition. The method can also be used in other objects recognition. Experimental result shows the approach can improve face recognize rate of small sample efficiently.