Facial expression recognition under a wide range of head poses

Most of the facial expression recognition methods assume frontal or near-frontal head poses and usually their accuracy strongly decreases when tested with non-frontal poses. Training a 2D pose-specific classifier for a large number of discrete poses can be time consuming due to the need of many samples per pose. On the other hand, 2D and 3D view-point independent approaches are usually not robust to very large head rotations. In this paper we transform the problem of facial expression recognition under large head rotations into a missing data classification problem. 3D data of the face are projected onto a head pose invariant 2D representation and in this projection the only difference between poses is due to self-occlusions with respect to the depth sensor's position. Once projected, the visible part of the face is split in overlapping patches which are input to independent local classifiers and a voting scheme gives the final output. Experimental results on common benchmarks show tha...