loading subjects...
Facial expression recognition (FER) has attracted persistently more and more attention due to its wide application potentials and scientific challenges. In this paper, we propose a novel approach to 2D+3D FER using orthogonal low rank Tucker decomposition (OLRTDFER). First, a new 4D tensor is built by stacking nine kinds of feature from 2D textured images and 3D face scans. Then, under a Tucker decomposition of this tensor, the low-rankness is imposed on the involved core tensor due to the high similarity of samples during projecting the three-dimensional face scans into the two-dimensional planes. Meanwhile the sparse representation of the factor matrix involved is carried out to avoid its denseness. Finally, a tensor completion is then embedded because the information is partly missed in the process of generating this 4D tensor. The validation performance are carried out on the BU-3DFE database, and the competitive results are obtained.