三维颅骨特征点的自动标定  

Automatic feature point extraction for three-dimensional skull

作　　者：冯筠[1] 陈雨[1] 仝鑫龙 贺小伟[1] 周明全[2] 

机构地区：[1]西北大学信息科学与技术学院,陕西西安710127 [2]北京师范大学信息科学与技术学院,北京100875

出　　处：《光学精密工程》2014年第5期1388-1394,共7页Optics and Precision Engineering

基　　金：教育部虚拟现实应用工程研究中心(北京师范大学)2012年度开放基金资助项目;陕西省科技计划资助项目(科技新星No.2012KJXX-29);陕西省教育厅科学研究计划资助项目(No.12JK0729)

摘　　要：提出了颅骨特征点的全自动标定方法,该方法利用分区统计可变模型及模型相似性匹配的方法来标定颅骨特征点。首先,对颅骨分区样本进行统计建模;利用统计模型的形变控制生成基准模型和生成模型,并建立基准模型和生成模型间的映射关系。然后,定义了模型之间相似性。最后,利用模型相似度和映射关系,间接得到待测模型的特征点。实验结果表明:该方法定位眼眶模型特征点的位置平均误差值为3.232 5pixel;当距离阈值为10pixel(模型大小的3%)时,有90%的特征点的位置准确率达到100%。与现有方法相比,本文方法标定的颅骨特征点的准确度和精确度都更高,并且可以标定颅骨模型平滑区域的特征点。A fully automatic skull feature point extraction method was proposed,which extracts the skull feature points by apartitioned statistical deformable model and a model similarity matching method.First,the statistic models of skull partition were constructed,and a benchmark model and a series of generated models were built by statistical model deformation.Then the mapping relationship between models was established and the model similarity was defined.Finally,the feature points of the model to be measured were indirectly obtained with the model similarity and the projection relationship.Experimental results indicate that the location average error of the feature points for an eye socket model is about 3.232 5pixels.When the distance threshold is 10pixels（3% of the size of the model）,the location accuracy for 90%of the feature points achieves 100%.The method proposed has higher accuracy and exaction for skull feature point extraction as compared with traditional methods, and can extract the feature points of smooth regions for skull models.

关 键 词：医学图像 特征点标定 统计可变形模型 相似性匹配 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]