We introduce multilinear dimension-reduction and classification methods for video image sequences. Tensor-to-tensor projection methods for spatio-temporal data are derived as dimension-reduction methods using the three-mode tensor representation. The tensor-to-tensor projection methods transform a tensor to a product of smaller tensors. Furthermore, we construct efficient and robust multiclass classifiers for multilinear forms by using tensorial expressions of spatio-temporal video sequences.
展开▼
