Head yaw estimation via symmetry of regions

This paper proposes a novel method to estimate the head yaw rotations using the symmetry of regions. We argue and reveal that the symmetry between the two regions in the same horizontal row are closely relevant to the yaw rotation of head, while at the same time insensitive to the identity of the face. The proposed method relies on the effective combination of Gabor features and covariance descriptors. Specifically, we first extract the Gabor features of a face image, then the covariance descriptors are used to compute the symmetry of Gabor features. Since the covariance matrix can eliminate the influence which is caused by rotations and illuminations, the proposed method is robust to these variations. In addition, the proposed method can be further improved by combining it with supervised learning. Experiments on two challenging databases are conducted, on which the proposed method improves the current state-of-the-art.