Automatic 3D face landmark localization based on 3D vector field analysis

In applications such as 3D face synthesis and animation, a prominent face landmark is required to enable 3D face normalization, pose correction, 3D face recognition and reconstruction. Due to variations in facial expressions, automatic 3D face landmark localization remains a challenge. Nose tip is one of the salient landmarks in a human face. In this paper, a novel nose tip localization technique is proposed. In the proposed approach, the rotation of the 3D vector field is analyzed for robust and efficient nose tip localization. The proposed technique has the following three characteristics: (1) it does not require any training; (2) it does not rely on any particular model; (3) it is very efficient, requiring an average time of only 1.9s for nose tip detection. We tested the proposed technique on BU3DFE and Shrec'10 datasets. Experimental results show that the proposed technique is robust to variations in facial expressions, achieving a 100% detection rate on these publicly available 3D face datasets.