基于模板的秦俑三维碎片部位标注技术研究

发布时间：2018-04-17 13:49

本文选题：文物复原 + 三维模型分割　；参考：《西北大学》2015年博士论文

【摘要】：秦俑考古发掘得到的实测三维数据维度高,邻接关系复杂。大量高维复杂碎片自动拼接属NP难问题。本文围绕目标模型部位模板获取,应用部位模板对大量三维碎片进行部位标注展开研究,使得部位模板作为智能拼接的目标导向,将大量复杂邻接碎片按部位分类标注,从而缩小碎片试拼接范围,提高碎片拼接效率,降低拼接难度。主要的研究工作和创新点如下：(1)单模型部位模板获取：针对传统模型分割算法在整个模型上计算曲率时间复杂度高、不适合高维数据的问题,提出了基于区域曲率约束的三维模型分割线提取算法。利用约束区域减少离散曲率的计算范围,避免了过分割,提高了高维度数据单模型分割的效率。应用于单个主体分支明显的兵马俑模型上,能够得到“有意义”的分割部位,形成准确的部位模板。(2)模型集部位模板获取：对于复杂模型单模型分割算法难以产生一致性的分割结果。提出了基于多特征谱聚类的模型集一致性分割算法。算法利用多特征描述子产生多个相似性矩阵,自动计算多个相似性矩阵的权重,发挥多特征的优势实现无监督的模型集聚类分割,得到整个模型集的部位模板,克服了一致性分割算法准确度不高的问题。(3)基于表面特征相似度检索的碎片部位标注：提出了基于精确欧氏局部敏感哈希(E~2LSH)的相似碎片快速标注算法。首先提取碎片的表面特征,并对表面特征进行基于E~2LSH的哈希索引。其次对部位模板进行碎片特征同尺度随机采样,提取采样的表面特征作为已标注样本。最后基于E~2LSH算法检索出与已标注样本相似的碎片进行部位标注。基于E~2LSH的检索算法将检索的时间复杂度降到次线性,与基于k-means聚类的检索算法比较,证明了算法的高效性和精确性,对大规模高维数据具有良好的实用性。(4)秦俑三维模型语义网构建：提出了基于本体的兵马俑数字三维模型语义网构建方法。构建兵马俑领域本体,并基于WordNet对本体进行扩展,利用语义网检索相似碎片或者部位模板。实现对兵马俑领域知识及拼接关系的有效管理。
[Abstract]:The measured three-dimensional data obtained from archaeological excavations of terracotta warriors are of high dimensionality and complex contiguous relationship.A large number of high dimensional complex fragments are NP-hard problems.In this paper, a large number of 3D fragments are annotated by the part template, which can be used as the target orientation of intelligent stitching, and a large number of complex adjacent fragments can be labeled according to the location.In order to reduce the range of fragment splicing, improve the efficiency of splicing, reduce the difficulty of splicing.The main research work and innovation are as follows: (1) single model location template acquisition: aiming at the problem that the traditional model segmentation algorithm has high time complexity in computing curvature on the whole model, it is not suitable for high-dimensional data.An algorithm based on region curvature constraint is proposed to extract the Secant line of 3D model.The constrained region is used to reduce the computational range of discrete curvature, to avoid over-segmentation, and to improve the efficiency of high-dimensional data sheet model segmentation.When applied to a model of terracotta warriors with a distinct branch of a single main body, a "meaningful" segmentation can be obtained.To form an accurate part template, it is difficult to generate consistent segmentation results for a single model segmentation algorithm for complex models.A model set consistent segmentation algorithm based on multi-feature spectrum clustering is proposed.The algorithm uses multiple feature descriptors to generate multiple similarity matrices, automatically calculates the weights of multiple similarity matrices, realizes unsupervised model clustering class segmentation, and obtains the part template of the whole model set.This paper overcomes the problem that the accuracy of the consistent segmentation algorithm is not high. (3) debris location tagging based on the similarity retrieval of surface features: a fast algorithm based on accurate Euclidean local sensitive hashing (EH2LSH) is proposed.Firstly, the surface feature of the fragment is extracted, and the surface feature is indexed based on E~2LSH.Secondly, the debris feature of the site template is sampled randomly at the same scale, and the surface feature of the sample is extracted as the labeled sample.Finally, the fragments similar to the labeled samples are retrieved based on the E~2LSH algorithm for site tagging.The retrieval algorithm based on E~2LSH reduces the time complexity to sublinearity. Compared with the retrieval algorithm based on k-means clustering, the algorithm is proved to be efficient and accurate.Construction of 3D model semantic web for terracotta warriors and terracotta warriors: a method of constructing digital 3D model semantic web of terracotta warriors based on ontology is proposed.The domain ontology of Terracotta Warriors is constructed, and the ontology is extended based on WordNet, and similar fragments or site templates are retrieved by semantic web.To achieve effective management of Terracotta Warriors and Terracotta Warriors in the field of knowledge and splicing relationship.
【学位授予单位】：西北大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：K876.3;TP391.41

【参考文献】