Three - Dimensional Model Retrieval Based On Spin Graph

With the constant improvement of multimedia technology and the maturing of digital geometry processing theory, the modern information technology industry has been developing rapidly. As an important product, three-dimensional model have been widely used in the fields of pattern recognition, computer-aided design and so on. However, there is a diversity of the expressional forms for three-dimensional model, which results in the retrieval and the identification becoming a problem for three-dimensional model, therefore, how to efficiently and accurately find the three-dimensional models which meet the demands becomes an important research topic in the three-dimensional retrieval field.Currently, the retrieval methods of the three-dimensional model can be roughly divided into two forms: one is based on the text and the other is based on the content, wherein, the content-based retrieval method can extract the shape features of three-dimensional model by analyzing and classifying the shape properties of the three-dimensional model so as to effectively solve the problem of retrieving. This article focuses on the analysis and exploration of the several retrieval methods which regard the shape feature of three-dimensional model as the signature, and then elaborates the advantages and disadvantages of these retrieval methods. Through analyzing and concluding the existing retrieval methods, we propose a feature descriptor of the three-dimensional models which is robust to rotation, translation and scale. The main researches are as follows:(1) We set the spin figure parameter and get average spin image descriptor based on the relative position between the local vertices.(2) We analyze the spin image feature descriptor of interference and verify which is robust to rotation, noise and point cloud density.(3) Finally, we select an appropriate matching recognition criterion to ensure to establish optimal matching relations between the two models.Experimental results show that the spin image shape descriptor compared with the other two have a good retrieval performance in both precision and recall rate evaluation, and can be applied for 3d model retrieval.