利用曲率多特征融合改进人体三维脊椎模型特征点的标注方法

发布时间：2018-05-28 01:18

  本文选题：脊椎 + 三维模型　； 参考：《光学精密工程》2016年11期

【摘要】：由于脊椎生理结构的精准坐标描述和准确匹配尚未达到医学精度的要求,本文对如何精确描述脊柱腰骶段特征点的物理坐标进行研究。介绍了人体脊椎采样特征点的定义和传统标记方法。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曲率多特征融合的自适应标注特征点的方法。该方法首先找出某个特征点的高斯曲率和平均曲率流的定义值,得到该特征点的法曲率相对极大值,并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。由于极大值曲率越大,三维模型表面在该点处的弯曲程度越大,该点就越能表现三维模型的大致轮廓,故以r范围内极大值最大的点作为特征点的曲率描述来替换手动拾取的点,从而准确反映该点的特征变化情况。最后,对改进结果进行偏差验证分析。结果表明:改进方法的准确度比现有手动标注特征点方法的准确度提高了约37%,验证了本文方法的有效性。
[Abstract]:Since the precise coordinate description and exact matching of spinal physiological structure have not reached the requirement of medical accuracy, this paper studies how to accurately describe the physical coordinates of the lumbosacral feature points of the spine. This paper introduces the definition and traditional marking method of human spine sampling feature points. In order to solve the problem that the accuracy of manual feature points is not enough and the errors are easy to be generated, a method of adaptive feature point tagging based on multi-feature fusion of curvature is proposed. In this method, the defined values of Gao Si curvature and mean curvature flow of a certain feature point are first found, and the relative maximum of normal curvature of the feature point is obtained, and the relative maximum of normal curvature of all model points in the specified minimum radius r is calculated. Because the greater the maximum curvature, the greater the curvature of the surface of the 3D model at this point, and the more the point can represent the approximate outline of the 3D model. Therefore, the point with the maximum value in r range is used as the curvature description of the feature point to replace the point picked up manually, so as to accurately reflect the characteristic change of the point. Finally, the deviation analysis of the improved results is carried out. The results show that the accuracy of the improved method is about 37% higher than that of the existing manual marking method, and the effectiveness of the proposed method is verified.
【作者单位】： 西北工业大学计算机学院;西北工业大学软件与微电子学院;西安电子科技大学计算机学院;
【基金】：国家自然科学基金资助项目(No.61172147)
【分类号】：TP391.41;R681.5
，

本文编号：1944628