一种基于法曲率极大值和向量内积的脊椎光学模型特征点自动识别方法  

An Automatic Recognition Algorithm for Feature Points of Spine Optical Model——Based on Maximum Value of the Curvature and Vector Inner Product

作　　者：惠宇[1] 武君胜[2] 杜静[3] 鱼滨[4] 张琛[4] 聂文斌 

机构地区：[1]西北工业大学计算机学院,陕西西安710072 [2]西北工业大学软件与微电子学院,陕西西安710072 [3]西北工业大学管理学院,陕西西安710072 [4]西安电子科技大学计算机学院,陕西西安710071 [5]西安点云生物科技有限公司,陕西西安710077

出　　处：《西北工业大学学报》2017年第6期1064-1072,共9页Journal of Northwestern Polytechnical University

基　　金：国家自然科学基金(61172147)资助

摘　　要：在脊椎光学模型的定位和配准中,主要难点是模型特征点的定位。针对手动标注特征点精确度不够,易产生较大误差等问题,提出了一种基于法曲率极大值和向量内积的脊椎模型特征点自动识别方法,该方法可以动态调整拾取点曲率,从而最大限度地保证特征点拾取的精准性。方法首先通过高斯曲率和平均曲率流等多曲率特征,得到手动选取点的法曲率极大值,由于曲率越大,三维模型表面在该点处的弯曲程度也就越大,即就是更能表现三维模型的几何轮廓信息;并以手动拾取点为圆心,计算在指定极小半径r范围内的所有模型点的法曲率相对极大值,然后对这些法曲率极大值进行降序排序,筛选出法曲率极大值较大的n个候选点,候选点与手动选取点之间做向量内积,从而得到向量之间的夹角。由于向量内积之间几何夹角角度越小,代表2个点之间欧几里得距离越靠近,故以夹角最小的候选顶点来替换手动拾取的点,从而可以准确反映该点的局部特征变化情况。经过实验的对比分析,新方法对特征点的标记准确性提高了约35%,从而验证了新方法的有效性。In the localization and registration of the spine model, the main difficulty is the location of the model feature points. In order to make the 3D model of human vertebrae more accurate, this paper proposes an automatic recognition algorithm for feature points of the spine model based on the Maximum value of the curvature and vector inner product, which can dynamically adjust the curvature of the pick-up point, so as to ensure the accuracy of feature points＇ pick-up to the greatest degree. Firstly, the curvature maxima of a manually selected point are obtained by using the multi-curvature features such as Gaussian curvature and mean curvature flow, and the relative maxima of normal curvature of all model points within the specified minimum radius r are calculated. The Maximum value of the curvature are sorted in descending order, and n candidate points with larger Maximum value of the curvature are selected. The vector inner product between the candidate points and the manually selected points is chosen to obtain the angle between vectors. The higher the curvature is, the greater the bending degree of the 3D model surface at that point is, that is, it is the better representation of the geometric contour information of the 3D model, and the smaller the geometric angle between the vector inner products is, representing the Euclidean distance between the two points is closer, so the manually picked points are replaced by the candidate vertices with the smallest angle, which can reflect the local characteristics of the point accurately. After comparing and analyzing the experimental results, the new algorithm improves the tagging accuracy of the feature points by about 35%, which verifies the effectiveness of the proposed algorithm.

关 键 词：三维光学模型 特征点标注 法曲率极大值 向量内积 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]