Estimating facial pose from sparse representation (MA-L2)

Author(s) : Hankyu Moon (NEC Labs America, USA) Matt Miller (NEC Labs America, USA)

Abstract : We present an approach to accurately estimate the pose of the human head in natural scenes. The essential features for estimating the head pose are the positions of the prominent facial features relative to the position of the head. We have developed a high-dimensional, randomly sparse representation of human face using a simplified facial feature model. The representation transforms a raw face image into sets of vectors representing the fits of the face to the random, sparse set of model configurations. The transformation is designed to collect salient features of the face image that is useful to estimate the pose, while suppressing any irrelevant ariations of face appearance. The relation between the sparse representation and the pose is learned using the SVR (Support Vector Regression). The sparse representation combined with the SVR learning is shown to estimate the pose sufficiently well.

Menu