Automatic Facial Feature Correspondence Based on Pose Estimation

Establishing facial feature correspondence across image of the same subject with different poses is an essential issue in the field of face image interpretation. Traditional approaches involve tedious landmark labeling and time-consuming training process. Instead, an automatic facial feature correspondence method is proposed in the paper, which achieves accurate feature correspondence via sparse feature matching using unary and geometric constraints, and intelligent interpolation based on facial pose angle estimation. The experiments show the validity of the proposed method.