基于局部重构与全局保持的半监督判别分析方法

为克服线性判别分析(LDA)只能利用有标记样本的缺点,提出一种基于局部重构与全局保持的半监督判别分析(LRGPSSDA)方法.LRGPSSDA通过最小化局部重构误差来确定邻域图的边权值,在保持数据集局部结构的同时保持其全局结构,具有对邻域参数的选择不敏感、所得投影子空间的维数不受样本类别数的限制等特点.相较现有的半监督判别分析方法(如SDA和UDA),LRGPSSDA的分类性能更好.在YaleB和CMUPIE标准人脸库上的实验结果验证了该算法的有效性.

Semi-Supervised Discriminant Analysis Method Based on Local Reconstruction and Global Preserving

Considering the shortcoming of Linear Discriminant Analysis (LDA) that it can only make use of labeled samples but can’t make use of large amount of unlabeled samples, a new method of Local Reconstruction and Global Preserving Based Semi-Supervised Discriminant Analysis (LRGPSSDA) is proposed in this paper. Compared with existing semi-supervised discriminative analysis methods, LRGPSSDA can preserve the global geometric structure of the sampled data set when preserving its local geometric structure and can set the edge weights of neighborhood graph through minimizing the local reconstruction error, besides it isn’t sensitive to the selection of neighborhood parameter and the dimensionality of its low dimensional projecting subspace doesn’t suffer from the restriction of the number of classes. The experimental results on YaleB and CMU PIE face database demonstrate the effectiveness of the proposed method.