An Efficient Partial Shape Matching Algorithm for 3D Tooth Recognition

Abstract

As a new biometric strategy, tooth recognition has drawn much attention in recent years. However, most existing work focus mainly on 2D dental radiographs which are less informative and vulnerable to noise and pose variance. Although there are already several attempts on 3D tooth recognition, the results are still inaccurate and performance is inefficient. Moreover, existing methods cannot recognize precisely when the post-mortem data contains incomplete teeth. In this work, we propose an efficient and accurate partial shape matching algorithm to recognize 3D teeth for human identification. Given the ante-mortem and post-mortem teeth models which were taken from patients using a laser scanner, we first extract a series of stable and consistent feature points on the surface of 3D teeth models using a sparse feature selection method based on the saliency map. For each feature point we then establish descriptor based on Improved Spin Images (ISI), which is able to accurately describe the local region around the feature points. Due to the small number of feature points, their correspondences can be efficiently found via the ISI descriptors. Finally, the similarity of the teeth of two input samples (ante-mortem and post-mortem data) can be determined by the sum of the distances between the corresponding ISI descriptors of the feature points. We also conduct experiments to show that the proposed method can achieve state-ofart performance for both complete and incomplete postmortem teeth data.

Keywords