3D face recognition using local shape map

This paper firstly proposes a new scheme for description of 3D local shape, termed local shape map (LSM). The LSM for a point on the surface is a 2D histogram constructed from mapping 3D coordinates of surface´s points within a sphere centralized at this point into a 2D space. The similarity metric for LSM is given. Secondly, application of LSM to 3D face recognition is presented, which is performed by incorporating LSM into an ad-hoc voting method. It does not require registering two face models. With a dataset of 31 facial range images for six subjects, in presence of significant change in pose and slight variation in expression, the equal error rate of 2.98% is obtained. The experimental results also demonstrate the performance of 3D face recognition using LSM outperforms that using spin image.