基于局部多尺度的三维模型草图检索方法

发布时间：2018-08-13 18:21

【摘要】：三维模型检索已经成为虚拟现实领域的基本问题,草图是三维模型检索的最好方式,随着各种不同分辨率三维模型数据的快速增长,现有基于草图三维模型检索算法无法有效应用于大规模复杂三维模型数据库的检索.在该文中,为了解决低分辨率三维模型顶点数和面片数较少,不易获取完整光滑几何线图的问题,提出了利用高斯差分方法对三维模型深度缓存投影图像提取三维模型线图,并采用贝塞尔曲线优化线图结果.为了提高检索准确率,提出了一种局部多尺度多方向的特征计算方法,在珈伯多方向滤波的基础上,建立了局部形状金字塔尺度空间结构,配合梯度方向直方图计算局部形状特征描述,并利用词袋方法组织局部形状特征.为了提高检索速度,利用加权索引方法存储归一化后的图像特征向量,实现基于草图的三维模型实时检索过程.该文不仅比较分析了不同参数下方法的执行效果,而且运用多种评价方法,选用3个不同三维模型数据库进行方法测试,实现检索原型系统验证方法的有效性.实验证明,该方法相比于大部分其他方法具有更好的检索效果.
[Abstract]:3D model retrieval has become a basic problem in the field of virtual reality. Sketch is the best way to retrieve 3D model. The existing Sketch-based 3D model retrieval algorithms can not be effectively applied to the retrieval of large-scale complex 3D model databases. In this paper, in order to solve the problem that the number of vertices and facets of low resolution 3D model is less and it is difficult to obtain the complete smooth geometric graph, a Gao Si difference method is proposed to extract the 3D model graph from the depth cache projection image of 3D model. Bessel curve was used to optimize the results. In order to improve the retrieval accuracy, a local multi-scale multi-directional feature calculation method is proposed. Based on the Gabor multi-direction filtering, the local shape pyramid scale spatial structure is established. The local shape feature description is calculated with gradient histogram, and the local shape feature is organized by word bag method. In order to improve the retrieval speed, the normalized image feature vector is stored by weighted index method, and the real-time retrieval process of 3D model based on sketch is realized. This paper not only compares and analyzes the execution effect of the method under different parameters, but also uses a variety of evaluation methods, selects three different 3D model databases to test the method, and realizes the validity of the verification method of the retrieval prototype system. Experiments show that this method has better retrieval effect than most other methods.
【作者单位】：北京师范大学信息科学与技术学院;北京师范大学虚拟现实应用教育部工程研究中心;浙江大学文物与博物馆学系;
【基金】：国家自然科学基金(61001168,61202198) 中央高校基本科研业务费专项资金(2013YB67)资助~~
【分类号】：TP391.41

【相似文献】