3D Face Recognition Using Local Features Matching on Sphere Depth Representation.

This paper proposes a 3D face recognition approach using sphere depth image, which is robust to pose variations in unconstrained environments. The input 3D face point clouds is first transformed into sphere depth images, and then represented as a 3DLBP image to enhance the distinctiveness of smooth and similar facial depth images. An improved SIFT algorithm is applied in the following matching process. The improved SIFT algorithm employs the learning to rank approach to select the keypoints with higher stability and repeatability instead of manually rule-based method used by the original SIFT algorithm. The proposed face recognition method is evaluated on CASIA 3D face database. And the experimental results show our approach has superior performance than many existing methods for 3D face recognition and handles pose variations quite well.