一种应用于人脸识别的有监督NMF算法

A Supervised Non-Negative Matrix Factorization Algorithm for Face Recognition

为了提高非负矩阵分解(NMF)算法识别率,提出了一种有监督的NMF(SNMF)方法。该算法对NMF基图像进行判别分析,然后选择主要反应类内差异的基图像来构造子空间,最后在子空间上进行识别。通过UMIST人脸库和CMUPIE人脸库上的实验结果表明,该方法对光照、姿态和表情变化具有一定的鲁棒性,识别率高于NMF方法和其它子空间分析法。

A supervised NMF algorithm to enhance the classification accuracy of the NMF algorithm is presented. The method employs discriminant analysis in the features derived from NMF. In this way, intrasubject variation is minimized, while the intersubject variation is maximized feature extraction procedure. Experimental results on public available face databases show that the proposed method has higher recognition rate than NMF and other subspace methods.