三维模型的局部匹配和检索方法研究
发布时间：2018-12-14 03:41
【摘要】：随着三维模型技术和互联网技术的不断发展，越来越多的三维模型制作软件和三维模型文件被共享在互联网上面，同时三维模型技术的应用领域也越来越广泛，如产品设计、制造仿真、虚拟现实、3D网络游戏等。尤其是最近几年3D打印机的出现，三维模型的应用已经开始普及至家庭用户，使得家庭用户可以应用3D打印机打印自己所需的三维模型。因此，研究和开发三维模型搜索引擎帮助企业用户、家庭用户快速、准确的检索到自己所需的三维模型，是最近几年的研究热点之一。 论文针对三维模型特征描述和检索这一问题展开研究，主要工作包括以下几个方面： 研究了基于形状统计方法的特征提取算法，针对传统的D2形状描述算法的不足，提出了一种基于面积算子的三维模型检索算法，该算法首先将对三角网格模型进行模型简化处理，使得表达三维网格模型的点和面的集合达到最小化，然后对三维网格模型上的顶点进行统计分析，对每个顶点所关联的三角形的面积进行计算，并对点相关的面积进行归一化处理，然后对点关联的面积序列进行通傅里叶变换，得到特征向量，做出面积分布图，通过计算模型间特征向量的差异，得到模型间的差异，进而检索出相似的三维模型。实验表明，和类似的基于形状统计的模型检索算法相比较，该方法可以获得更好的检索结果。 针对三维模型检索中的局部特征描述和匹配问题，提出了一种基于顶点邻域属性的三维模型检索算法，该算法首先统计三维网格模型的相关属性和顶点的邻域顶点相关属性，包括邻域的质心、顶点到邻域质心的矢量、顶点到邻域质心的距离、顶点的法向矢量、顶点的曲率以及顶点的法向矢量和顶点邻域质心矢量之间的夹角等，接着对邻域顶点的属性中的2个角度进行16等分，形成一个16×16规模的特征矩阵，然后采用一组特定的矩阵相似度计算方法计算特征矩阵间的相似度，，最后计算模型整体的相似度，以此来代替二个三维模型的相似度。通过和其他检索算法的比较实验表明，该方法对具有较丰富局部特征的模型可以获得更好的检索结果。 提出了一种基于法向夹角直方图的三维模型检索算法。算法首先对三维模型进行预处理，接着定义了三维模型上的三角网格各个顶点的法向与三角网格之间的夹角计算方法，然后根据三角网格三个顶点的法向与三角网格之间的夹角，对三角网格进行分类，根据夹角是锐角还是钝角，将三角网格分成四种类型，对每种类型的三角网格集合构造形状分布曲线，通过对模型的四条形状分布曲线的比较，得出两个模型的相似度，从而实现模型的相似性检索。实验表明，该算法的检索准确率和检索效率方面要由于其他类似的直方图算法。 在三维模型的局部匹配和检索方面，引入了三维分割技术，首先采用基于Laplace-Beltrami算子特征函数的模型特征描述方法，采用有限元算子方法离散化和求解Laplace-Beltrami算子特征值问题，将算子特征函数值作为三维模型的特征描述符；然后采用K均值聚类方法，对特征描述符进行聚类分析，聚类后的结果将三维模型分割成多个局部区域；最后采用指派问题中的匈牙利算法计算三维网格模型之间的局部区域与整体模型之间的匹配度，从而得到模型间的匹配结果。 为了验证前面提到的算法，设计了一个三维模型检索系统的原型系统，系统中提供了多种检索算法，包括面积分布算子算法、顶点邻域属性算法和基于LB算子特征函数的算法以及课题组其它成员设计的算法。
[Abstract]:With the development of the three-dimensional model technology and the Internet technology, more and more three-dimensional model production software and three-dimensional model files are shared on the Internet, and the application field of the three-dimensional model technology is also more and more widely, such as product design, manufacturing simulation, virtual reality, and the like. In particular, the appearance of the 3D printer in recent years, the application of the three-dimensional model has started to spread to the home user, so that the home user can use the 3D printer to print the three-dimensional model required by the user. Therefore, the research and development of the three-dimensional model search engine to help the enterprise users, home users to search quickly and accurately to the three-dimensional model that they need, is one of the research hot spots in recent years. This paper studies the characteristics of the three-dimensional model and the retrieval of this problem. The main work includes the following: In this paper, the feature extraction algorithm based on the shape statistic method is studied, and the traditional D2 shape description algorithm is used. In this paper, a three-dimensional model retrieval algorithm based on the area operator is proposed. The algorithm first will model the triangular mesh model, so that the set of points and faces expressing the three-dimensional mesh model is minimized, and then the vertexes on the three-dimensional mesh model are unified. The method comprises the following steps of: calculating an area of a triangle which is associated with each vertex, carrying out normalization processing on the area related to a point, performing a Fourier transform on the area sequence associated with the point, obtaining a feature vector, the difference between the models is obtained to obtain the difference between the models, and then the similar three are retrieved. The experimental results show that the method can be better tested compared with similar model retrieval algorithms based on shape statistics. As a result, a three-dimensional model retrieval algorithm based on the vertex-neighborhood attribute is proposed for the local feature description and the matching problem in the three-dimensional model retrieval. The algorithm first counts the related attributes of the three-dimensional mesh model and the adjacent vertex-related attributes of the vertex. the centroid of the neighborhood, the vector of the vertex to the centroid of the neighborhood, the distance from the vertex to the centroid of the neighborhood, the method of the vertex, the curvature of the vertex and the angle between the method of the vertex and the centroid vector of the vertex, and then the two angles in the attribute of the adjacent vertex Line 16, etc., form a 16-16-16-scale feature matrix, and then calculate the similarity between the feature matrices using a set of specific matrix similarity calculation methods, and finally calculate the similarity of the whole model to replace the two three-dimensional models. The similarity of the model can be obtained by comparison with other search algorithms. The three-dimensional angle histogram based on the method is presented in this paper. The method comprises the following steps of: pre-processing a three-dimensional model, and then defining an included angle calculation method between the method of each vertex of the triangular mesh on the three-dimensional model and the triangular mesh, The angular mesh is classified, and the triangular mesh is divided into four types according to the acute angle or the obtuse angle, and the shape distribution curve is constructed for each type of triangular mesh set. By comparison of the four shape distribution curves of the model, the similarity of the two models is obtained, and the model is realized. The result of the experiment shows that the retrieval accuracy and the retrieval efficiency of the algorithm are due to other classes In terms of the local matching and retrieval of the three-dimensional model, the three-dimensional segmentation technique is introduced. First, the model feature description method based on the characteristic function of the Laplace-Beltrami operator is adopted. The method of finite element operator is used to discretize and solve the Laplace-Beltra. the characteristic function value of the operator is used as the characteristic descriptor of the three-dimensional model, the K-means clustering method is adopted, the characteristic descriptor is subjected to cluster analysis, and the result of the clustering is the three-dimensional model and finally, calculating the matching degree between the local area and the whole model between the three-dimensional mesh model by using the Hungarian algorithm in the assignment problem, In order to verify the above-mentioned algorithm, a prototype system of a three-dimensional model retrieval system is designed, and a variety of search algorithms are provided in the system including the area distribution operator algorithm, the vertex neighborhood attribute algorithm and the algorithm based on the characteristic function of the LB operator and the class.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP391.41
[Abstract]:With the development of the three-dimensional model technology and the Internet technology, more and more three-dimensional model production software and three-dimensional model files are shared on the Internet, and the application field of the three-dimensional model technology is also more and more widely, such as product design, manufacturing simulation, virtual reality, and the like. In particular, the appearance of the 3D printer in recent years, the application of the three-dimensional model has started to spread to the home user, so that the home user can use the 3D printer to print the three-dimensional model required by the user. Therefore, the research and development of the three-dimensional model search engine to help the enterprise users, home users to search quickly and accurately to the three-dimensional model that they need, is one of the research hot spots in recent years. This paper studies the characteristics of the three-dimensional model and the retrieval of this problem. The main work includes the following: In this paper, the feature extraction algorithm based on the shape statistic method is studied, and the traditional D2 shape description algorithm is used. In this paper, a three-dimensional model retrieval algorithm based on the area operator is proposed. The algorithm first will model the triangular mesh model, so that the set of points and faces expressing the three-dimensional mesh model is minimized, and then the vertexes on the three-dimensional mesh model are unified. The method comprises the following steps of: calculating an area of a triangle which is associated with each vertex, carrying out normalization processing on the area related to a point, performing a Fourier transform on the area sequence associated with the point, obtaining a feature vector, the difference between the models is obtained to obtain the difference between the models, and then the similar three are retrieved. The experimental results show that the method can be better tested compared with similar model retrieval algorithms based on shape statistics. As a result, a three-dimensional model retrieval algorithm based on the vertex-neighborhood attribute is proposed for the local feature description and the matching problem in the three-dimensional model retrieval. The algorithm first counts the related attributes of the three-dimensional mesh model and the adjacent vertex-related attributes of the vertex. the centroid of the neighborhood, the vector of the vertex to the centroid of the neighborhood, the distance from the vertex to the centroid of the neighborhood, the method of the vertex, the curvature of the vertex and the angle between the method of the vertex and the centroid vector of the vertex, and then the two angles in the attribute of the adjacent vertex Line 16, etc., form a 16-16-16-scale feature matrix, and then calculate the similarity between the feature matrices using a set of specific matrix similarity calculation methods, and finally calculate the similarity of the whole model to replace the two three-dimensional models. The similarity of the model can be obtained by comparison with other search algorithms. The three-dimensional angle histogram based on the method is presented in this paper. The method comprises the following steps of: pre-processing a three-dimensional model, and then defining an included angle calculation method between the method of each vertex of the triangular mesh on the three-dimensional model and the triangular mesh, The angular mesh is classified, and the triangular mesh is divided into four types according to the acute angle or the obtuse angle, and the shape distribution curve is constructed for each type of triangular mesh set. By comparison of the four shape distribution curves of the model, the similarity of the two models is obtained, and the model is realized. The result of the experiment shows that the retrieval accuracy and the retrieval efficiency of the algorithm are due to other classes In terms of the local matching and retrieval of the three-dimensional model, the three-dimensional segmentation technique is introduced. First, the model feature description method based on the characteristic function of the Laplace-Beltrami operator is adopted. The method of finite element operator is used to discretize and solve the Laplace-Beltra. the characteristic function value of the operator is used as the characteristic descriptor of the three-dimensional model, the K-means clustering method is adopted, the characteristic descriptor is subjected to cluster analysis, and the result of the clustering is the three-dimensional model and finally, calculating the matching degree between the local area and the whole model between the three-dimensional mesh model by using the Hungarian algorithm in the assignment problem, In order to verify the above-mentioned algorithm, a prototype system of a three-dimensional model retrieval system is designed, and a variety of search algorithms are provided in the system including the area distribution operator algorithm, the vertex neighborhood attribute algorithm and the algorithm based on the characteristic function of the LB operator and the class.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP391.41
【参考文献】
相关期刊论文 前10条
1 张玉鹏;周明全;李晖;;基于形状相似的三维模型检索方法研究[J];北京师范大学学报(自然科学版);2011年04期
2 胡敏;罗s
本文编号：2377850
本文链接：https://www.wllwen.com/kejilunwen/sousuoyinqinglunwen/2377850.html
