Multi-pose face recognition combining tensor face and manifold learning

This paper proposes a method of multi-view face recognition based on pose estimation combining tensor face and manifold learning effectively. It firstly calculates the nearest pose of the test sample in the training set with LEA (Locally Embedding Analysis), and reduces the tensor´s view dimension. Then it uses LLE (Locally Linear Embedding) to reduce the tensor´s pixel dimension. Finally, It uses the method of tensor face and the pose of the test sample to do face recognition. This method has resolved the problem of lack of nonlinearity that exists in tensor face presentation. The experiments on the face databases of FacePix, Weizmann show that the novel method significantly improves the accuracy of original tensor face recognition.