加入迭代因子的层次化颅骨配准方法

发布时间：2018-01-22 02:57

本文关键词： 颅骨配准特征区域积分不变量主成分分析法改进迭代最近点(ICP)算法迭代因子　出处：《中国图象图形学报》2017年04期 　论文类型：期刊论文

【摘要】：目的在基于知识的颅面复原中,为了对未知颅骨的面貌进行复原,需要在颅骨库里寻找相似颅骨,将相似颅骨的面皮作为参考。寻找相似颅骨的过程即颅骨配准,配准的精度和效率是两个重要性能指标。本文提出一种基于特征区域和改进ICP(iterative closest point)算法的层次化颅骨配准方法。方法首先,将颅骨模型去噪、简化并归一化,通过计算体积积分不变量,确定每个点的凹凸性;使用K-means方法,将颅骨上的点聚类为多个或凹或凸的特征区域。然后,通过主成分分析法来计算两个颅骨的相似特征区域,对每一个可能的匹配计算3维变换,将两个颅骨粗略对齐;最后,采用加入迭代因子的方法对ICP算法进行改进,使用改进的ICP算法对颅骨进行精配准。结果将本文方法用于颅骨模型、兵马俑模型以及公共数据集中的3维模型配准,经典ICP算法的配准时间分别为6.23 s、7.61 s、4.17 s,改进的ICP算法配准时间分别为3.02 s、3.23 s、2.83 s,算法效率提高了约2倍,配准效果也有明显提高。实验中通过对迭代因子的测试,发现不同的数据集需要设定不同的迭代因子。结论本文所提出的基于区域特征的层次化配准方法提高了颅骨配准的精度和效率,整个过程不需要人工干预,该算法具有一定的普适性,可用于相似3维模型配准。
[Abstract]:Objective in the process of knowledge based craniofacial restoration, in order to recover the unknown skull, we need to look for the similar skull in the skull bank. The face skin of the similar skull is taken as reference. The process of finding the similar skull is the registration of the skull. The accuracy and efficiency of registration are two important performance indexes. In this paper, a new method based on feature region and improved ICP(iterative closest point is proposed. The hierarchical skull registration method of the algorithm. The skull model is de-noised, simplified and normalized, and the convexity of each point is determined by calculating the volume integral invariant. Using K-means method, the points on the skull are clustered into multiple concave or convex feature areas. Then, the similar feature areas of the two skulls are calculated by principal component analysis. Three dimensional transformation is calculated for each possible match, and the two skulls are roughly aligned. Finally, the ICP algorithm is improved by adding iterative factors, and the improved ICP algorithm is applied to the fine registration of the skull. Results the method is applied to the skull model. For the terracotta Warriors model and the 3D model in the common data set, the registration time of the classical ICP algorithm is 6.23 s-1 7.61 sm 4.17 s respectively. The registration time of the improved ICP algorithm is 3.02 s-1 3.23 s-1 2.83 s, and the efficiency of the improved algorithm is increased about 2 times. The registration effect is also improved obviously. The iterative factor is tested in the experiment. It is found that different data sets need to set different iterative factors. Conclusion the hierarchical registration method based on region features proposed in this paper improves the accuracy and efficiency of skull registration without manual intervention. The algorithm is universal and can be used for similar 3D model registration.
【作者单位】：西北大学信息科学与技术学院;
【基金】：国家自然科学基金项目(61305032,61572400,61373117) 陕西省教育厅科研专项(2013JK1180)~~
【分类号】：TP311.13;D919
【正文快照】： Vol.22,No.4,April 2017ent skulls is subsequently calculated by comparing their principal components and areas.Optimal matching is conductedthrough exhaustive search.The optimal 3D transform for each potential pair of matched feature regions approximately

【相似文献】