face identification that based on tensor SVD decomposition

Abstract: By using an approach based on the SVD(Tensor Singular Value Decomposition) in the extraction and expression of human face features in the process of face recognition, the precedent algorithms, such as the PCA(Principal Component Analysis) which has excessive dependence on the condition of face pictures, are improved. The SVD method deals with three-dimensional linear data model, which can avoid the decrease of precision caused by the variation of picture conditions when using method deals with two-dimensional linear data model, and provides a relatively stable result despite the change of conditions. In addition, by using QR decomposition of matrix to reduce the complexity of calculation without jeopardizing the accuracy, the algorithm is optimized efficiently. Four groups of experiments based on Matlab are conducted, and the results are analyzed in comparison with those from the PCA method, which verifies the outstanding correctness and stability of the algorithm under varying conditions. Meanwhile, experiments on the optimized algorithm show a remarkable improvement of efficiency compared to the basic algorithm, especially when the data amount gets larger.

The recognition result of my algorithm using yale A database.We can see most people's photo are correctly math

Result of orl database

Result of my friends' photo