Fast and Reliable Facial Landmarks Localization in Non Frontal Images

The goal of this paper is to introduce a pipeline able to precisely and rapidly localize a set of facial landmarks on a face image. The proposed method is also able to estimate the pose of the face portrayed on the image independently from its orientation and it is robust to partial occlusions. This method is based on an optimized version of Part Mixture Model, it is able to fit a facial landmarks configuration to faces dealing with out-of-plane rotations and is suitable for real time processing. Moreover, the proposed method also uses a double scale approach to speed up the pose estimation phase and a posteriori configuration regularization based on Principal Component Analysis (PCA).