Range Face Image Registration Using ERFI from 3D Images

摘要

In this paper, we present a novel and robust approach for 3D faces registration based on Energy Range Face Image (ERFI). ERFI is the frontal face model for the individual people from the database. It can be considered as a mean frontal range face image for each person. Thus, the total energy of the frontal range face images has been preserved by ERFI. For registration purpose, an interesting point or a land mark, which is the nose tip (or 'pronasal') from face surface is extracted. Then, this landmark is exploited to correct the oriented faces by applying the 3D geometrical rotation technique with respect to the ERFI model for registration purpose. During the error calculation phase, Manhattan distance metric between the localized 'pronasal' landmark on face image and that of ERFI model is determined on Euclidian space. The accuracy is quantified with selection of cut-points 'T' on measured Manhattan distances along yaw, pitch and roll. The proposed method has been tested on Frav3D database and achieved 82.5% accurate pose registration.