DBSCAN和K-Means混合聚类的牙齿特征自动识别

DBSCAN and K-Means Hybrid Clustering Based Automatic Dental Feature Detection

根据牙齿网格数据表面的特点,提出一种密度聚类DBSCAN和K-Means混合聚类的牙齿特征自动识别算法.首先利用平均曲率和高度值对数据进行预处理,加大区域间的距离;然后将牙齿点中高度值高的点投影到XOY平面作DBSCAN聚类,获取簇数和中心点作为下一步的输入;再使用K-Means算法对预处理的模型处理,用于牙齿区域的划分;最后基于每个划分区域对特征点进行识别.实验结果表明,该算法能够精确地检测出牙齿的特征点,比已有识别算法操作简单、正确率高.

According to the characteristics of dental meshes, this paper proposed an automatic dental feature detection algorithm based on DBSCAN（density-based spatial clustering of applications with noise） and K-Means hybrid clustering. First, average curvatures and geodesic distances were used to better represent the distances between partition. Second, DBSCAN was applied to projective points which have larger Z values in order to obtain the number of clusters and center points as the input of next step. Next, we used K-Means algorithm to generate partitions of the dental mesh. Finally, the feature detection algorithm was employed to obtain feature points on each partition. Experiments show that this algorithm can accurately detect dental features. Compared to the previous automatic algorithm, the accuracy is significantly improved and the operations are also simplified.