高斯核非负矩阵因子及其在表情识别中的应用

摘要

提出一种基于高斯核非负矩阵因子的人脸表情识别方法.该算法引入高斯核函数并结合NMF(Non-negative Matrix Factorization)进行表情特征提取,称之为GKNMF.与NMF、2DNMF(2-Dimensional Non-negative Matrix Factorization)等方法不同,GKNMF通过基于高斯核的非线性映射可从原始表情数据中提取更多线性和非线性的有用信息,尽可能地保留原始的表情信息.根据JAFFE和CED-WYU(1.0)两个表情数据库的识别结果表明,GKNMF特征提取方法能有效地提高识别率.%A facial expression recognition method based on Gaussian kernel non-negative matrix factor is proposed in this paper. The algorithm introduces Gaussian kernel function and combines NMF to extract facial expression, so it is named GKNMF. Unlike NMF and 2DNMF, through Gaussian kernel-based nonlinear mapping, GKNMF can extract more linear and non-linear useful expression features hidden in original expression image and remain original expression information as much as possible. Recognition results gained from the expression library of CED-WYU (1.0) and JAFFE show that the GKNMF is an effective method for improving the recognition accuracy.