Selfie to Avatar - Creating a 3D Model of a Face from Front and Side View Photos

Abstract: In this thesis we present a system for creating a textured, animated 3D model of a user’s face in real time. The model, which we call an avatar, is reconstructed from two photos of a subject, one from a front view and one from a side view. Each photo is segmented using Active Shape Models. The shapes determined by the segmentation are used to reconstruct a 3D model and the pixel colors of the images are used to texture it. In order to evaluate the system, we use a test where test respondents try to match real world photos of people to their corresponding avatars. Although we do not take any special care of modelling hair, which is a difficult problem, we achieve a recognition rate of 82.7 % which demonstrates that our method already can provide realistic avatars of a person with little input from the user.