Face alignment via an ensemble of random ferns
Xiang Xu, Shishir K. Shah, Ioannis A. Kakadiaris · 2016
This paper proposes a simple but efficient shape regression method for face alignment using an ensemble of random ferns. First, a classification method is used to obtain several mean shapes for initialization. Second, an ensemble of local random ferns is learned based on the correlation between the projected regression targets and local pixel-difference matrix for each landmark. Third, the ensemble of random ferns is used to generate local binary features. Finally, the global projection matrix is learned based on concatenated binary features using ridge regression. The results demonstrate that the proposed method is efficient and accurate when compared with the state-of-the-art face alignment methods and achieve the best performance on LFPW and Helen datasets.