基于线性判别局部保留映射的人脸表情识别

Facial Expression Recognition Using Discriminant Locality Preserving Projections

随着人机交互技术的发展,情感计算成为一个研究热点。局部保留映射(LPP)是一种最优的保持数据集局部结构的一种线性映射,它的特点是保留了样本的局部结构,但是它没有考虑判别信息,从而容易引起类间距离小的类别之间的重叠。本文提出了基于线性判别的局部保留映射(DLPP)算法并将其应用到表情识别问题中。与LPP相比,DLPP的改进之处在于将判别分析的思想引入LPP。同时考虑样本间的相邻关系和模式类之间的相邻关系,从而得到能正确分类的最优投影方向。在Yale人脸库和JAFFE表情库中的一系列表情识别实验结果表明,DLPP对于人脸表情识别更为有效。

Due to the development of human-computer interaction,there has been much interest in facial recognition. Locality Preserving Projections （LPP） is a linear projection that arises by solving a variational problem that optimally preserves the neighborhood structure of the data set. LPP is designed for preserving local structure, but it fails to take account of the discriminant information. Thus, Discriminant Locality Preserving Projections （DLPP） is presented in this paper, and it＇ 11 be applied to facial expression recognition problem. The improvement of DLPP algorithm over LPP is that discriminant analysis has been introduced into LPP algorithm. By consid- ering both the neighborhood structure of the samples and the neighborhood of the classes ,we obtain an optimal projecting direction which can classify the samples more efficiently. A series of experiments were performed on Yale and JAFFE face databases and the result shows that DLPP is more efficient for facial expression recognition problem.