3D Face Registration using Individual Reference Face Shape

3D face biometrics has become an important research topic that has seen an enormous growth in the last few decades. The performance of any 3D face recognition system depends on the accurate registration or alignment of facial surfaces. Various registration methods in literature so far use either the one-to-all registration approaches or the average face model (AFM) based approaches for aligning the probe 3D face to gallery faces. In the former approach, each probe face is registered to all faces in the gallery, at a great computational cost. In the latter approach, the AFM calculated from the gallery images is considered as the reference face for aligning all the probe faces. However, the alignment of probe face to the less similar AFM results in the possible loss of discriminatory information. In this paper, an alternative approach that uses individual reference shape for the registration of each probe 3D face is proposed. Here each probe face is aligned to its corresponding, much similar reference face shape that ensures a better registration. Experimental results show that the proposed registration approach provides better alignment than the existing methods and the results are reported using the Bosphorus 3D face database.