一种新颖的基于NMF的人脸识别方法

Novel NMF-based Face Recognition Method

提出了一种新颖的基于非负矩阵分解(NMF)进行人脸识别的方法,该方法首先对人脸图像作NMF得到由人脸基图像组成的子空间矩阵及其对应的系数矩阵,然后将其重构获得NMF重构人脸图像,进而从原始图像中减去重构图像以获取残差图像。最后对残差图像作局部非负矩阵分解(LNMF)以提取其非负子空间及其系数矩阵,并将两次得到的系数矩阵混合,根据最近邻原则进行识别。在ORL标准人脸数据库上的实验表明,该方法具有更好的识别率。

A novel NMF-based method for face recognition was put forward. First, NMF was applied to the face images for extracting non-negative subspaee and the corresponding coeffeient matrices, then they were reconstructed to obtain the NMF reconstructed images. Moreover, the residual images were computed by subtracting reconstructed images from original face images. Finally, the residual images were applied by LNMF to extract non-negative subspace and the corresponding coefficient matrices, the two coefficent matrices were combined for face recognition based on the nearest neighbor principle. The simulation experiments illustrate that this method has better recognition rate on the ORL face database.