基于统计学的三维模型检索算法

【作者】 周刚；

【导师】 郭鹏江；

【作者基本信息】 西北大学 ， 应用数学， 2009， 硕士

【摘要】 三维模型在虚拟手术、分子生物、文物保护、计算机辅助设计等众多领域都扮演着非常重要的角色。上个世纪90年代以来,随着这些领域的快速发展以及网络三维模型海量数据库的扩充和三维物体扫描技术的成熟,导致了三维模型数量的急剧增加,人们迫切需要从众多的三维模型中准确地找到自己所需要的模型,因此三维模型检索越来越受到研究人员的重视。本文在对前人关于三维模型检索的工作进行系统研究的基础之上,提出了两种新的基于统计学的三维模型检索算法。本文所做的工作如下:(1)总结与分析了三维模型检索的研究现状、相关技术。(2)实现了两个前人的基于统计学的三维模型检索算法:Osada提出的形状分布的检索算法和Ankerst提出的对模型切分后生成直方图的检索算法。(3)提出了一种新的基于统计学的三维模型检索算法:基于相对角度直方图匹配的检索算法。该算法仅仅通过统计模型表面点上每一个点与其他所有点之间的角度关系作为模型的特征量用于模型的检索,不需要考虑模型三角面片的相关信息。(4)由于相对角度直方图需要考虑模型上每一个点与其他所有点之间的角度关系,导致了每一个模型所形成的直方图的维数较大,所以基于降维的考虑又提出了一种基于聚类分析的检索算法。该算法大大的降低了模型的特征量维数,从而降低了算法的时间复杂度,加快了模型检索的速度。(5)总结了算法性能评价的相关标准,并且从实验结果与算法性能来看,本文提出的两种三维模型的检索算法对于大多数模型来说在检索效果与算法性能上都比Osada提出的形状分布的检索算法和Ankerst提出的对模型切分后生成直方图的检索算法要好。更多还原

【Abstract】 3D model retrieval plays a very important role in so many fields such as virtual surgery, molecular biology, heritage protection, computer aided design. Since the 90’s, with the rapid development of these fields as well as the expansion of massive database of web 3D model and the scanning technology maturity of three-dimensional objects, the quantities of 3D models are sharply increased, it’s urgent need for people to accurately find their own required model from a large number of 3D models. So 3D model retrieval becomes more and more importance to researchers. Based on the systematic study about the previous work of 3D model retrieval, this paper presents two new kinds of 3D model retrieval algorithm based on statistics. The work done of this paper is as follows:(1) The 3D model retrieval’s research status and related technologies are summarized and analyzed.(2) Two previous 3D model retrieval algorithms based on statistics are realized: the algorithm of shape distribution which was presented by Osada and the algorithm of the model cutting generated histogram which was presented by Ankerst.(3) A new 3D model retrieval algorithm based on statistics is presented: the retrieval algorithm based on the relative angle histogram matching. The algorithm simply need to statistic the angle’s relationship of each model point of model’s surface point with all other points as model’s features for the model retrieval, it needs not to consider the model’s related information of triangle.(4) A new retrieval algorithm of cluster analysis based on dimension reduction in this paper is presented as the histogram of relative angle needs to considerate angle’s relation of all points, which has lead to large dimension of histogram generated for each model. This algorithm greatly reduces the characteristic index dimension of model, thus reducing algorithm time complexity and accelerating the speed of model retrieval.(5) The relevant standards of the algorithm performance evaluation are concluded, and from the experimental results and the algorithm performance, two kinds of 3D model retrieval algorithm which are presented by this paper have better than the algorithm of shape distribution which was presented by Osada and the algorithm of the model cutting generated histogram which was presented by Ankerstfor the most models.更多还原