摘要
二维主成分分析方法是直接利用二维图像来构建方差矩阵的。为了充分利用样本类别信息,文章以类间散布矩阵特征向量作为投影方向进行特征抽取。首先用2DPCA先作一次横向压缩,对抽取出的特征矩阵再用2DPCA作一次纵向压缩。与传统二维主成分算法比较,极大压缩了特征的维数,加快了分类速度,提高了识别率。用ORL人脸数据库进行了实验验证,证明了本方法的可行性。
Two-dimensional PCA is to build the variance matrix directly using two-dimensional image matrix. In order to fully use the sample category information, the eigenvectors of the between-class scatter matrix are used as the projection direction to extract features in this paper. First the image is compressed in horizon direction using 2DPCA, and then the extracted feature matrix is compressed again in vertical direction. Compared to the traditional two-dimensional PCA, the dimensions of the features are greatly reduced, the speed of classification is accelerated, and the recognition rate is improved. The experiment results based on 0RL face database indicate that the method is feasible.
出处
《电子技术（上海）》
2012年第3期23-24,22,共3页
Electronic Technology
