Résumé: How to effectively reduce the dimensionality of high-dimensional and massive unconstrained face data is important. In this paper, the Weighted Tensor Sparse Graph Embedding (WBTSGE) algorithm is proposed. The original sample image is divided into B blocks, and each image block is represented by a second-order tensor. Then, the algorithm introduces the sample category label and added the intra-class compactness constraint. The sparse reconstruction coefficient is constrained by adding distance weight. Finally, the optimal transformation matrix is obtained by adding global constraint factors. The experimental results show that the recognition rate of WBTSGE is 98.57% in the AR face database and 99.80% in the Extended Yale B database. The WBTSGE algorithm improves the accuracy of face recognition in the experimental face database collected in the controlled environment and the real face database collected in the uncontrolled environment.

Langue: Anglais