A two-stage facial landmark localization method

Facial expression analysis plays an important role in nonverbal immersive communications. Considering facial landmark localization is an essential part for facial expression analysis, it draws much attention. Since the accuracy of facial landmark localization is related with different expressions and is affected by the result of face detection, we propose a two-stage facial landmark localization method in this paper. First, a method which models face detection, facial alignment and expression recognition simultaneously is used to roughly locate the facial features/landmarks. Then a supervised descent method is used to refine the locations of facial features. Experimental results on the extend Cohn-Kanade database demonstrate that the proposed method has more accurate facial landmarks localization result.