关键词: 人脸识别, 加权小波变换, 非负矩阵分解, 流形正则化

Abstract: In order to improve the recognition performance by obtaining more sufficient face features, the method of weighted wavelet decomposition and manifold regularized non-negative matrix factorization is introduced to realize face recognition. Firstly, wavelet decomposition with its weighted high frequency is applied to extract the features of weighted high frequency component and low frequency component from training samples. Secondly, with maintaining potential geometric structures and local features of the face features, it uses manifold regularized non-negative matrix factorization to acquire final recognition characteristics. Lastly, nearest neighbor method is used to be classified and recognized. Comparing with the traditional method of non-negative matrix factorization, experimental results on ORL face databases and YALE face databases show that the recognition rate is about increased by 5% and computation time is quite shorter. Hence, the proposed method has less time consuming, as well as a better recognition performance.

Key words: face recognition, weighted wavelet decomposition, non-negative matrix factorization, manifold regularization