二阶分块PCA的人脸特征提取方法研究

为了提取更为有效的鉴别特征,在已有的二阶特征脸方法和分块主成分分析(PCA)方法上,提出了二阶分块PCA人脸特征提取方法.该方法对原始人脸图像和经重建得到的剩余图像分别运用分块PCA,将提取的一阶和二阶特征线性组合为一个特征矩阵,再进行分类识别.此特征能更充分反映人脸图像的低频和高频特性.采用ORL人脸库和FERET人脸库的实验结果表明该二阶分块PCA正确识别率优于普通分块PCA算法,具有较强的特征提取能力.

Face Feature Extraction Method with Second-Order Modular PCA

A face feature extraction method based on second-order modular PCA(principal component analysis) is proposed to capture more effective discriminative features under the second-order eigenface method and the modular PCA method.A feature matrix can be obtained through a linear combination of first-order feature and second-order feature extracted by the method that applies modular PCA to original face image and remnant image which is reconstructed.This feature can more fully reflect the low-frequency and high-frequency characteristics of face image.Experimental results based on the ORL face database and FERET face database indicate that recognition accuracy of this second-order modular PCA is higher than common modular PCA algorithm and the proposed method has a strong feature extraction capability.