高级检索

3D图形匹配中多对称模型的新算法

New Algorithm for Multi-symmetry Models in 3D Shape Matching

  • 摘要: 3D图形匹配在计算机视觉领域有着广泛的应用, 其中对称图形因其几何特征十分相似, 难以区分, 其匹配问题一直是难点之一. 本文主要以多节肢动物模型为研究对象, 提出一种基于几何特征的多对称图形匹配的新算法. 主要步骤为: 1. 选取特征点, 即在热核信号极值点的基础上, 采用最远点采样法和融合算法对特征点个数进行调整, 得到特征点集; 2. 对特征点进行分类, 引入对称差异度和支持点对的概念将特征点分为对称点和非对称点, 再利用测地距离将对称点进一步分侧, 以提高后期的匹配准确率; 3. 图形匹配, 首先通过算法在非对称点集中确定一个参考点, 利用对称点与该参考点的距离排序完成初始匹配; 针对可能出现的左右交叉错误问题, 对初始匹配结果进行调整, 即确定模型的正方向, 通过判断对称点与参考点所成向量的外积方向与正方向是否一致, 将交叉错误的匹配结果进行矫正, 最终得到正确的匹配结果. 将本文算法在TOSCA数据库中进行实验, 结果表明, 针对蚂蚁、蜘蛛模型, 无论是在对称点检测上, 还是在匹配结果上, 本文算法与已有算法相比, 其正确率和运行效率均有所提高.

     

    Abstract: Three-dimensional shape matching has a wide range of applications in the field of computer vision, among which symmetric shape matching has always been one of the difficult problems because its geometric features are very similar and difficult to distinguish. This thesis is focused on the matching problem of 3D multi-symmetric shape. In this thesis, a new algorithm of multi-symmetric shape matching is proposed with a multi-arthropod model as the research object. The main steps are: First, select the feature points, that is, based on the extreme value point of the Heat Kernel Signature, the farthest point sampling and fusion algorithm is used to adjust the number of feature points to obtain the feature point set; Second, classify the feature points, the concept of symmetry difference and supporting point pairs is introduced to divide feature points into symmetric points and asymmetric points, and then geodesic distance is used to further divide the symmetric points to improve the matching accuracy in the later stage; Third, conduct the shape matching, the algorithm is given to determine a reference point in the set of asymmetric points, using the distance ordering of the symmetry point from the reference point to complete the initial matching; In order to adjust the possible left-right cross error problem, the normal of shape is determined, and by judging whether the cross product direction of the vector formed by the symmetry point and the reference point is consistent with the normal to obtain the correct matching result. The experimental results of this algorithm in TOSCA database show that the correctness and operation efficiency of this algorithm are improved compared with the existing algorithms for the ant and spider models, both in symmetric point detection and in matching results.

     

/

返回文章
返回