Abstract: An improved regional statistical craniofacial model algorithm is proposed for mass unequally spaced point sets from CT images and the smoothness in the seems of the eyes, nose, mouth regions. In this paper, the covering spheres with two different radii are used to sample the points from CT images, so a craniofacial model is decomposed into two regions：T region（eyes, nose, mouth region）and O region（the other region）. Moreover, the information of race, gender, age, bmi is added to represent the craniofacial model. The statistical craniofacial model is constructed using PCA. The improved algorithm has showed great effects compared to original algorithm especially to smoothness of the craniofacial model.

Key words: craniofacial reconstruction, Principal Component Analysis（PCA）, statistical craniofacial model, landmark, CT image

摘要： 针对初始颅面点集数据量大、分布不均匀，以及分区统计颅面模型算法处理颅面眼、鼻、口三个区域不易连接的问题，提出了改进的分区统计颅面模型算法。通过将颅面分成T区（眼、鼻、口区域）和O区（其他区域），解决了眼、鼻、口区域的连接问题，并用两种不同半径的球覆盖颅面的方法处理初始颅面点集，使颅面两个区域内点集分布均匀。同时引入种族、性别、年龄、胖瘦四个属性表示单个样本模型，利用主成分分析方法建立统计颅面模型。实验结果表明该算法能够取得更好的颅面重构效果。

关键词: 颅面重构, 主成分分析, 统计颅面模型, 特征点, CT图像