Sparse feature extraction for model-less robust face pose estimation

Head pose estimation has been a focus of research in computer vision both implicitly in tasks that use full body pose estimation, and explicitly in tasks that perform face tracking and face recognition. One of the main problems in the field of face recognition techniques is the difficulty of handling varying poses. In this work, we propose a robust pose estimation approach that represents the relationship between the features of the face and the corresponding labels through a sparse representation. The discrimination of this representation is enhanced by projecting to a new space, pushing similar samples closer to each other. Experiment results are provided to show the robustness and accuracy of our approach.