Ye Pan and Luciano Silva
3D and Range Data Analysis, 3D Face Indexing, Iterative Closest Point, Simulated Annealing
In this paper we present a novel 3D face indexing and recognition framework. First, 3D facial scans are aligned with a 3D facial template by applying a modified Iterative Closest Point (ICP) algorithm and the distance vector of 3D facial regions are computed to generate the indexes of gallery. For recognition we select k-nearest candidates according to the indexes and use a Simulated Annealing based registration approach to match a probe with candidate faces. Our experimental results on the Face Recognition Grand Challenge (FRGC) v2 database show that our indexing approach is effective and could eliminate approximately 80% matches with 1% recognition rate loss, which could yield 98% rank one recognition performance at 0.001 False Acceptance Rate(FAR). Our results were compare very favorably to the ones from published state-of-the-art methods.
Important Links:
Go Back
