基于测地回归模型的颅面复原  

Craniofacial Reconstruction Based on Geodesic Regression Model

作　　者：陈仲晗 赵俊莉 于晗 潘振宽 贾彬 李劲华 Chen Zhonghan;Zhao Junli;Yu Han;Pan Zhenkuan;Jia bin;Li Jinhua(School of Data Science and Software Engineering,Qingdao University,Qingdao 266071;College of Computer Science and Technology,National University of Defense Technology,Changsha 410073;College of Computer Science and Technology,Qingdao University,Qingdao 266071)

机构地区：[1]青岛大学数据科学与软件工程学院,青岛266071 [2]国防科技大学计算机学院,长沙410073 [3]青岛大学计算机科学技术学院,青岛266071

出　　处：《计算机辅助设计与图形学学报》2021年第3期395-404,共10页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61702293,61772294);全国统计科学研究项目(2020355);山东省重点研发计划重大科技创新工程(2019JZZY020101).

摘　　要：颅面复原是指根据一个未知颅骨的特征预测出对应的面貌,在考古研究、医学整容、刑事案件调查等领域有重要应用.为解决颅面复原过程中存在着数据量大、需要大量标定特征点的手工工作及颅面特征点定义困难的问题,针对三角网格表示的三维颅骨和面皮模型,将颅面模型用从鼻尖出发的一组测地线表示,提出了基于测地回归的颅面复原方法.该方法首先从三维人脸模型的鼻尖点提取一组测地线;然后以颅骨和对应的人脸模型上提取的测地线作为训练样本,采用偏最小二乘回归统计模型复原未知颅骨对应人脸模型的测地线;最后通过迭代最近点算法将生成的测地线与人脸统计模型进行匹配,实现复原测试颅骨的面貌.在192套颅面数据库上进行了颅面复原实验,将基于测地回归法复原的结果与主成分分析法和偏最小二乘回归法复原结果对比,实验结果表明,该方法能够提高复原精度,减少复原时间,取得了良好的复原效果.Craniofacial reconstruction refers to predicting the corresponding appearance based on the characteristics of an unknown skull.In order to solve the problem of large amount of data,manual work of calibrating feature points and difficulty in defining craniofacial feature points in the process of craniofacial reconstruction,for the three-dimensional skull and skin model represented by a triangular mesh,the craniofacial model is represented by a set of geodesics which are starting from the tip of a nose,thereby craniofacial reconstruction method based on geodesic regression is proposed.First,a set of geodesics is extracted from the nose tip point of the 3D face model;then the skull and the geodesics extracted from the corresponding face model are used as training samples,and the partial least square regression statistical model is used to reconstruct geodesics corresponding to the unknown skull;Finally,the iterative closest point algorithm is used to match the geodesics generated by the geodesic regression method with the facial statistical model to reconstruct the face of the test skull.Craniofacial reconstruction experiments have been conducted on 192 sets of craniofacial data.The reconstruction results based on geodesic regression were compared with the reconstruction results of principal component analysis and partial least squares regression.The experimental results showed that our method can improve the reconstruction accuracy,reduce the reconstruction time,and achieve a good reconstruction effect.

关 键 词：颅面复原 测地线 偏最小二乘回归 测地回归统计模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]