关键词: 颅骨配准, 神经网络, 迭代最近点, 尺度因子, 模拟退火

Abstract: Skull registration is an important step of craniofacial restoration,the accuracy of which plays a key role to the recovered results.To improve the registration accuracy of skulls and solve the registration problem of different solutions,a 3D skull point cloud registration method from coarse to fine is proposed in the paper.Firstly,coarse registration of skulls is completed using the point cloud registration method based on neural network (NN).Then the iterative closest point (ICP) is improved by integrating scale factor and simulated annealing coefficient into it and used to complete skull fine registration.The improved ICP algorithm could improve the accuracy and speed of fine registration and achieve the final registration of skulls with high accuracy.An unknown skull is registered with 300 skulls in database in the experiment,the results show that the NN-based point cloud registration method could align the skulls coarsely,and the registration accuracy and convergence rate of improved ICP algorithm increase by about 30% and 50% respectively compared with ICP algorithm.So the proposed skull registration method from coarse to fine is an effective skull registration method which could complete accurate registration of 3D skull model.

Key words: skull registration, neural network, iterative closest point, scale factor, simulated annealing