现有的三维颅面复原技术大多依据颅骨特征点的软组织厚度统计值。针对现有统计值指标涵盖的年龄、胖瘦等属性段较宽泛导致复原面貌缺乏个性的缺点，提出了一种改进方法。首先通过CT扫描仪获得颅面样本数据，并通过图像重构获得三维颅骨和人脸模型；然后采用一种半自动特征点标定方法对三维颅骨样本进行特征点标定，并求解特征点软组织厚度；之后采用支持向量回归方法构建特征点软组织厚度与属性之间的函数关系；最后根据待复原颅骨的属性以及回归函数计算特征点软组织厚度，在此基础上采用薄板样条函数对参考人脸模型进行变形获得复原面貌。实验结果表明，相比于已有方法，该方法能获得更准确的软组织厚度，提高颅面复原的准确度。%Most of the 3D craniofacial reconstruction methods rely on the statistical data of soft tissue depths of sparse land-marks located on the skull.The classical statistical method for tissue depth is to classify samples into several clusters accor-ding to the properties (gender,age and BMI)of the samples,and then calculates the mean tissue depths for each cluster. However,each cluster covers a wide range of properties,for example,and BMI,leading to a result that are insensitive to the slight changes of properties.This paper proposed an improved method to solve this problem.The method first constructed a head database from CT images,and located 80 landmarks for each skull of the database by using a semi-automatic landmarking method.Then,it calculated the tissue depths of the 80 landmarks for all the skulls,and analyzed the relationship between tis-sue depth and properties,such as gender,age and BMI,for each landmark through support vector regression.When recon-structing the face for a given skull,it first calculated the tissue depths of landmarks according to the regression function and the properties of the skull,and then deformed a reference face using thin-plate spline based deformation to obtain the approxima-tion face for the skull.The experiments demonstrate that the proposed method can get more sensitive and more accurate tissue depths for landmarks when comparing with the existing methods,so as to improve the accuracy of the reconstruction.
展开▼
