一种双向压缩的二维特征抽取算法及其应用

Two Dimension Double PCA for Extract Features and Application

针对二维主分量分析(2DPCA)在最优投影轴上的投影是一个向量,其抽取出的图像特征是一个矩阵,特征数据量大,不便于直接分类的弱点,提出了一种特征抽取新方法。首先用2DPCA作一次横向压缩,对抽取出的特征矩阵再作一次2DPCA进行纵向压缩。这样抽取出的特征数量大大减少,可加快分类速度。ORL人脸库的试验结果表明了该方法的有效性。

Because the features extracting by Two Dimension Principal Component Analysis （2DPCA） are matrixes, it needs much space to store these features and slow down the classification speed. We propose a novel feature extraction algorithm called Two Dimension Double PCA （2DDPCA） in this paper. First we use 2DPCA compressing the images in horizon direction, then we compress the features in vertical direction using 2DPCA again. Thus, the dimension of features is lesser and the speed of classification is faster. The experiment on ORL face database indicates that the proposed method outperform 2DPCA.