Face Recognition Using a Multilinear Framework

Appearance-based method classifies images by their appearances, using statistical knowledge learned from training data. This paper presents a framework of multilinear algebra-based face recognition, which proposes the use of tensor projector. We will discuss the choice of distance function, which is not usually paid attention to in most research on this area. In fact, a properly chosen distance function is nontrivial to the recognition rate, even for the linear Tensorface method. This implies that if we apply a more advanced method such as ICA, an even better performance can be achieved. At the end of this paper, we present our experiment result based on the framework, which was performed on the Freiburg face database.