Dense reconstruction of 3D human face using 5 images and no reference model

3D reconstruction of human faces is of high importance in applications like forensics, facial features based human tracking and realistic reconstruction of human faces in virtual reality applications. The published algorithms so far require either a large number of views of a human face or a 3D facial reference model. This paper proposes a technique for 3D human face reconstruction using only five views without any reference model unlike most of the contemporary facial reconstruction techniques. Face localization is done followed by facial feature point extraction in the facial images. The features are tracked in the other images using the Kanade-Lucas-Tomasi (KLT) object tracking algorithm. 3D location of each feature is calculated in all the images using triangulation and a 3D point cloud is formed. The Point cloud is processed to remove outliers and holes. The 3D face is then formed by interpolating and meshing the point cloud. This computationally efficient and low cost technique outperforms commercial and online available software and published algorithms.