A hybrid approach for face alignment

Abstract

Face alignment has been an indispensable procedure in face application. It is still a challenge to locate facial landmarks in unconstrained scene. In this paper, we propose an algorithm to perform accurately face alignment. A method based on human retinal information processing principle is proposed to enhance the images and remove the illumination noise. According to the image’s locality principle, the distance constrains is imposed on the pixel difference features around the landmark to achieve robustness. And then, the random forest is used to map the pixel difference features to local binary features. The obtained local binary features are used to jointly learn a linear regression for the final output. In addition, the inherent data structure is utilized to reduce the computational burden when preform maximum variance reduction on the split node in random forest. Extensive experiments on public datasets show that the proposed approach can locate facial landmarks accurately and rapidly.