A Survey On Content-based 3D Model Retrieval

With the development of computer software and hardware,3D model is in the number of the geometric growth,and 3D model is of the increasingly widespread use of research,such as entertainment, medicine, mechanical engineering, application fields etc. In addition, in biology, chemistry, geography and archeology fields, 3D data model has become increasingly popular reservoir. With the development of the Internet, people have good conditions of 3D model sharing and processing, retrieval of 3D model increasing people's attention. This was mainly due to :(1)Equipment for scanning three-dimensional model has improved the quality, it is convenient to scan high-quality 3D model.(2)Many 3-D modeling software, such as CAD, people can more easily design complex 3D models.(3)As the network's development, it could be widely disseminated on the Internet to exchange three-dimensional models conveniently .(4) Many fields are using a lot of 3D models, sometimes only partial revision of the existing model is needed.Retrieval of 3D model currently there are mainly two ways : One is the three-dimensional model based on text retrieval. Another is the 3-D content-based searches. Text retrieval based on the three-dimensional model was based primarily on three-dimensional model for the Notes search Some systems also three-dimensional model based on the size of the document or type of document retrieval, and the inadequacy of these methods is that they used only a few keys to the complex three-dimensional model, 3D scene description, do not fully .In addition, three-dimensional model of the Notes text keyword artificially added, which will inevitably be subjective and one-sidedness. poor reliability is obviously difficult to meet the people's demands .This will require people to come up with a new search method, content-based re- trieval methods. Based on the content of the 3D model is designed to reflect the use of visual features of the building features automatic indexing searched The purpose 3D model .Content-Based Retrieval of 3D visual features and the similarity matching to identify user requirements that the 3D model closer to the people in real life through the use of intuitive impression .Now,there are trillions of 3D models exist in the world, and every day there is a lot of 3D model generation .So, in the face of a huge database of three-dimensional model, the model of how to swiftly search to become a hot topic. It involves the artificial intelligence, computer vision, to identify areas of ambiguity .MPEG-7 has been in international standards, in addition to data related to the media, including two-dimensional media information, 3D also includes three-dimensional models and other virtual media messages.In this paper,I status the content-based retrieval technique of three-dimensional model and describe the progress of a broad and thorough review .First, the presentation and analysis of content-based retrieval techniques 3D model and the general retrieval processes are : standardization, three-dimensional model of feature extraction, feature matching, return search results :(I)Standardize the model. As the 3D model user profile varied, the model may be the same object model database in the course of their size and orientation is not the same .For example, the model is a 3D model databases tower's glass was replenished, stored in the database is a three-dimensional model of a tiny portion of the horizontally, originally with an object. It is only because of the different direction and placed a great impact on the search results, this need to find a way of making every direction towards a three-dimensional models .There are many ways, but the statistics is the most commonly used principal component analysis (PCA). This method can meet the above requirements, this paper presents a calculation of principal components analysis of the concrete steps .(II) Feature extraction .Feature extraction is most important in 3D model retrieval, the paper cited the three commonly used methods. They are all from different perspectives to the analysis of the 3D model.First,Based on the analysis of the geometric shape of the structure of feature extraction method .Such geometric method was to compare overall, a basic idea of an object A with another object is to place B overlap. The B objects calculated at objects outside the borders of the distance between points A to the distribution of objects, these include histograms constructed from the end to these histograms basis for the similarity metric. Such method can be applied to : stiffness, deformation, the closed mesh. The advantage is a simple principle : that human beings are similar with the standard of judgment for the overall match. Shortcomings of the model are : high standards and universal large amount of strict pretreatment requirements .Second,Based on the feature extraction method of topology. 3D model from the topology of such method, three-dimensional model of feature extraction, the main method of extracting a 3-D matrix, hyper-surfaces deformation regions, branches, skeleton etc.. Such method can be applied to : voxel model, the closed mesh. The advantages are : to satisfy the geometric invariance in the sense of visualization, for the part and the whole match. The shortcomings are : higher, or a branch in the appropriate section of the model calculation, the boundary can be affected by noise sensitively .Feature extraction method based on statistical characteristics of shape. Generally speaking, three-dimensional model to be extracted before the need for standardization of the model coordinates. The main method is principal component analysis (PCA), PCA has its inherent flaws and is not stable. Therefore, some researchers began seeking a three-dimensional model does not require a standardized model coordinates feature extraction method. Feature extraction based on statistical characteristics of the three-dimensional model based on Moment Method in addition to the geometric characteristics, There Legendre moments, Fourier-Mellin moments, psedo-Zernike moments and so forth. Among them, two-dimensional Zernike Moment Method is better than others in anti-noise, anti-redundancy and recognition. In theory, there was the 2D to 3D Zernike moments polynomials. 3D Zernike polynomials function in a three-dimensional model is the orthogonal projection function, not a simple formula. But also has a good retrieval performance. Such method can be applied to arbitrary mesh : (the model is also suitable for degradation). The advantages are: model calculation is not complicated and it is easy to understand, and most of the geometric is invariant. Most of the noise robustness of the border are for the overall match. Some weaknesses are : Feature unstable surface characteristics of the geometric characteristics of alternatives, the loss of some features topological characteristics of the sensitive part, the majority of features suitable for rough classification .Third,Based on the feature extraction topology became a very commonly used algorithm. This part includes some theorems and definitions .(III) Similarity matching. This section lists some commonly used for similarity matching distance measurement methods, such as Euclidean distance. Manhattan distance, and Hausdorff distance .(IV) Ways of inquiries and some key issues. In this part I status some 3D model retrieval system of inquiries, and list some main web sites on 3D model retrieval.