The research of facial features localization based on posterior probability deformable model

In depth research of the human face understanding, almost all of these are based on the extraction of facial features. Admittedly, it requires accurate localization and segmentation for facial features. In this paper, aiming at the problem that traditional method depends on the initial position in facial features localization, an improved algorithm based on deformable model is proposed. This algorithm applies probability distributions to strike the maximum posterior probability to improve the process of deformable model energy minimization. The performance of proposed algorithm is evaluated on Yale face databases. Experimental results show that the improved algorithm can overcome the defect that the model is sensitive to initial position. Further the improved algorithm effectively solves the robustness of facial features localization and improves the speed of localization.