Abstract: Traditional face recognition methods obtain universal subspaces by using all trained images. The subspace mainly represents the commonness of human faces with few sights of single person’s face. This paper presents a novel method named NMF-based image reconstruction for face recognition. It obtains the basis images by using each person’s pictures respectively and the features which are employed to reconstruct the images by mapping the test images to the basis images. The minimum reconstruction error is adopted to finish the facial recognition. The computer simulation in ORL face database illustrates that the method is effective.

Key words: Non-negative Matrix Factorization(NMF), face recognition, reconstruction, feature

摘要： 由传统的人脸识别方法产生的人脸特征子空间通常是由人脸库中所有训练样本产生的一个通用子空间，该空间更多地包含了所有人脸样本的共性特征，而忽略了个性特征。该文提出一种基于NMF图像重构的方法，以单个人的训练样本集获取其人脸特征子空间，将识别图像向每一个特征子空间中进行映射及重构，并以重构图像的误差作为判据实现人脸识别。在ORL标准人脸库进行的计算机仿真证实了该方法的有效性。

关键词: 非负矩阵分解, 人脸识别, 重构, 特征

CLC Number: 

• TP391.41