This paper presents a new skull point cloud registration method based on curvature maps to improve the registration accuracy and convergence speed of the skull point cloud model. First, a three-dimensional shape block centered on the feature points and containing its adjacent points is extracted from the skull point cloud, and all the points are projected onto the two-dimensional plane. Furthermore, the projection points are quantized into the corresponding units in the two-dimensional supporting area, and the weighted curvature is encoded as curvature distribution images to construct the region curvature map descriptors of the feature points. Then, matching point pairs are established by matching points with similar local shapes based on regional curvature map descriptors, and the rigid body transformation relationship between skull point clouds is calculated using the singular value decomposition method to realize skull coarse registration. Finally, the iterative closest point (ICP) algorithm is improved by introducing dynamic iteration coefficients and used to achieve fine skull registration. The experiment results demonstrate that the proposed rough registration method is an effective initial registration method. Compared with the original ICP algorithm, the improved ICP algorithm increases the registration accuracy and convergence speed by approximately 11% and 37%, respectively, and reduces the time-consumption by approximately 34%. The bunny point cloud model is used to verify the generalization ability of the proposed method. The results demonstrate that the registration effects of the improved ICP algorithm are better than those of the original ICP algorithm.