Human head reconstruction based on elastic deformable 3D model

Abstract

This paper proposes a framework for generating a photo-realistic 3D head model from one frontal and two side view pictures. First a set of feature points on the 3D general head model, including frontal contour and profile contour are selected; then their correspondences are found in the photo images. Based on these correspondence pairs, RBFN is adopted to deform the mesh of general head model to fit the human head. After that a panorama image construction method is used to generate a seamless texture by linearly blending images around the frontal head contour region. Lastly, this seamless texture is mapped to the head mesh to get a specific 3D head model.