三维模型检索中的语义类视图技术研究

项目名称： 三维模型检索中的语义类视图技术研究

项目编号： No.61272304

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 张三元

作者单位： 浙江大学

项目金额： 80万元

中文摘要： 随着三维模型在数量上的迅猛增长和在各个领域的广泛应用，如何开发搜索引擎帮助用户快速方便地查找设计所需的三维模型是急需解决的一个问题。包括草图在内的图像是三维模型搜索引擎中一种非常简单方便的查询接口。本课题研究支持图像接口的三维模型检索技术，其研究目标是解决如下两个问题：1）如何使不同姿态下得到的二维视图仍然具有相关性，显著提高检索查全率；2）如何采用更少的视图表示三维模型，显著提高检索效率。我们的研究思路是提取三维模型的姿态不敏感特征建立语义学习空间，形成语义类。对每个语义类中的三维模型，结合最优视图理论得到该语义类的代表性视图集合：语义类视图。而对用户提供的图像，只需要和语义类视图进行相似性评价就可以返回检索结果。和已有研究相比较，该方法不仅通过语义学习有效地提高了基于图像的三维模型检索查全率，而且减少的视图个数能够明显地提高检索效率。研究成果能应用于搜索引擎开发，产业价值也十分明显。

中文关键词： 基于图像的三维模型检索；语义类视图；骨架特征包直方图；动态视点；最优视图

英文摘要： Recent years, 3D models are not only greatly increasing in number, but also widely used in different product fields. In consequence, an urgent problem is how to help people to accurately and efficiently find their desirable 3D models by 3D searching engine. Besides sketch, image is a very simple and convenient query interface for retrieving 3D models. This proposal is focusing on an efficient and effective image based 3D retrieval by resolving two key problems: 1) greatly improve retrieving precision by making 2D views obtained from different poses be highly correlated; 2) greatly improve retrieving efficiency by using less views for 3D models. The main idea of the proposed method is to build semantic class by pose insensitive 3D shape features. For 3D models in each semantic class, some most representative views are extracted by combining best view and cluster analysis. In this way, the retrieving results can be obtained by comparing query image and views of semantic classes. Comparing to the previous researches, the proposed method greatly improve not only the precision of image based 3D retrieval, but also retrieving efficiency due to reduced views. The method is original in resolving image based 3D retrieval. Moreover,the work can be used for developing 3D searching engine.

英文关键词： Image based 3D model retrieval；views of semantic class；histogram of skeleton bag；dynamic view point；best view