Higher Order Orthogonal Iteration vs. Multilinear Principle Component Analysis for Tensor Subspace Learning

摘要

Tensor subspace learning has been of great interest in computer vision and pattern recognition applications. In this paper, we demonstrate the relationship Higher Order Orthogonal Iteration (HOOI) and Multilinear Principle Component Analysis (MPCA), which are two of the most fundamental and most popular techniques for tensor subspace learning. We also mathematically prove that MPCA is a special case of HOOI and investigate the performance of HOOR on face recognition.