An efficient approach of 3D ear recognition

The Iterative Closest Point(ICP) algorithm is usually used for 3D ear recognition in the literatures. However, the high computational cost of ICP limits the application of 3D ear biometrics. In this paper, we present an efficient approach based on local feature and ICP for 3D ear recognition. The local features are detected and represented with LSP(Local Surface Patch), and used to compute the initial transformation for matching with ICP. An elite preservation strategy is introduced to refine the candidate gallery ears that a modified ICP algorithm with kd-tree index, distance and uniqueness constraints is applied to. The proposed approach achieved a rank-1 recognition rate 98.55% on Collection J2 of UND biometrics datasets. Matching an ear with a gallery requires only 1.73 sec on average.