多流形判别分析在特征提取中的应用研究

Research and Application of Feature Extraction with Discriminative Multiple Manifold

为了更好地提高人脸识别率及其识别效率,提出了一种基于多流形判别分析(MMDA)的图像特征提取方法。在MMDA方法中,为了寻求能够同时最大化类间散布矩阵和最小化类内散布矩阵的判别矩阵,类间、类内分布图分别被用来描述类间和类内的分离性,类内图可以表示子流形的信息,而类间图可以代表多流形的信息,从而更好地实现分类。在ORL及FERET人脸数据库上进行实验,结果表明了MMDA方法在特征提取中的有效性。

To improve the face recognition accuracy and efficiency, a novel feature extraction method with discriminative analysis based multi-manifold （MMDA） is proposed in this paper. Distribution graph of intra-class and inter-class is described as the separation of intra-class and inter-class, respectively,aiming to find the discriminative metrix that maximize inter distribution and minimize the intra distribution. Graph of intra-class represents the sub-manifold, and graph of inter-class represents the multi-manifold so that it can separate classes better. Experiments on ORL and FERET show the efficiency of MMDA in feature extraction.