Discriminant Orthogonal Rank-One Tensor Projections for Face Recognition

Traditional face recognition algorithms are mostly based on vector space. These algorithms result in the curse of dimensionality and the small-size sample problem easily. In order to overcome these problems, a new discriminant orthogonal rank-one tensor projections algorithm is proposed. The algorithm with tensor representation projects tensor data into vector features in the orthogonal space using rank-one projections and improves the class separability with the discriminant constraint. Moreover, the algorithm employs the alternative iteration scheme instead of the heuristic algorithm and guarantees the orthogonality of rank-one projections. The experiments indicate that the algorithm proposed in the paper has better performance for face recognition.