To improve the accuracy of image classification

fully use the structural information of the data

and compress the image data

first

third order tensor image features are constructed.Then

non-negative matrix factorization (NMF) is used for dimension reduction.A method of choosing agood starting point is proposed using two-dimensional principal component analysis (2DPCA)

which uses the information of the image effectively.Next

in order to maintain the intrinsic structure of the manifold for tensor subspace

weighted matrix is derived according to the labels of images.Meanwhile the set of images is used to construct a fourth order tensor.A method for classification is proposed by using weighted high-order singular value decomposition for support tensor machine.The experimental results on two image databases show that the proposed can effectively improve the accuracy of image classification.