Abstract: Traditional algebraic feature extraction approachs are based on vector patterns. When patterns are not vectors such as facial images, these methods may meet many problems. Recently two-dimension PCA (2DPCA) can directly compute the features by using original image matrixes. But the features extracted by 2DPCA are also matrixes; it could cause the magnitude of features too much and slow down the classification speed. This paper combines the virtues of 2DPCA and PCA, presents a 2DPCA plus PCA feature extraction method. It firstly use 2DPCA to deal with the original image matrixes, and then uses PCA to compress the feature matrixes again. The experiments on ORL face database indicate that the 20 times average recognition rates are respectively PCA (94.98%), 2DPCA (95.48%) and 2DPCA+PCA (95.83%).

Key words: Feature extraction, Face recognition, Principal component analysis (PCA)

摘要： 传统的特征抽取算法是基于向量的，在模式是图像时并不方便。二维投影方法利用图像矩阵直接计算，虽然抽取特征速度快，但抽取出的特征是矩阵，对应的特征数量大，影响分类速度。该文结合二者的优点，先用二维投影处理原始图像，降维后再做主分量分析，抽取出少量的特征进行分类，识别率和分类速度均有提高。在ORL人脸库上20次实验的平均识别率达95.83%。

关键词: 特征抽取, 人脸识别, 主分量分析