Realization And Research Of Several Important Technologes Of 3D Models Retrieval Based Shape Feature

We live in a 3D world. The 3D model is a kind of natural & direct method of describing real world substances. Therefore it has a broad application in the areas like industrial design, games, virtual reality etc. The gradual popularization of three dimensional scanning & modeling techniques and related software has led to the production of large quantity of three-dimensional models. In order to fully use the resources and find their precise three-dimensional models, emphasis on the research of Three Dimensional Model Retrieval System has been increasing day by day. The derivation of the characteristics of three-dimensional model is one of the basic problems in the research of the Three Dimensional Model Retrieval System. The main content of this thesis is to improve & further describe some representative methods in derivation of characteristics of three-dimensional models.The first chapter introduced the background of shape feature extract of 3D model, with emphasis on the concept & characteristics of 3D models retrieval system, present state of three dimensional research of the form in China & World; trends in the development of 3D models retrieval technologies, the main work of the thesis, its structure & design; and characters & significance of the research.The respected knowledge on shape feature extract of 3D model is introduced in the second chapter. The first part describes the basic module of 3D models retrieval system based on form diagnosis; the second part introduces on the necessity of standardization & pretreatment of the three dimensional model, generally adopted methods and two forms of three dimensional method denotation–grid pattern & voxel pattern. The third part is devoted on drawing an comparative analysis of some shape feature extract techniques. The 4th, 5th & 6th parts describe some conditions that an ideal shape feature vector of 3D model should possess, the methods of measuring the similarity and measuring guidelines for generally used evaluation searches effect of the similarity of shape feature vector. And then introduced Princeton Shape Benchmark three-dimensional model database and three-dimensional model search testing database used in the experiments in this article.Chapter 3 to chapter 7 is the main body of this article.In the third chapter a fast voxelization algorithm research based on template matching method has been realized. The main point on increasing preciseness of three dimensional model searches is precisely deriving the basic characteristics of the given model. To derive characteristics of the given model precisely, first and foremost the analysis of characters of every voxel composing the given model is necessary. But different types of three dimensional model creating tools or three-dimensional model databases are surround polygons models that are not been voxelization, surround polygons. Thus the three-dimensional model voxelization algorithm plays an important role in three-dimensional search based on shape feature. The process of the three-dimensional model voxelization is divided into two stages in this article. First the surface of the model is voxelization using distance increment parts algorithm. Here the triangular surface of the model is converted into cubes of specific sizes, thus derive the characteristics of surface of the model. Then inner part of the model is voxelization to get the structural characteristics of the model. For the voxelization of inner part, first 6 three dimensional templates are designed and matched with every division of the inner part, thus the division that fits with template condition is taken as model for inner voxel.The fourth chapter improved the calculating efficiency of radial method of deriving three dimensional network model characteristics. D.V.Vrani first develops the method. The main idea of the method is parameterization of the surface of the sphere model and then emits sample rays evenly from the centre along the direction of particular latitude. These rays intersect with the surface of the model, and greatest value of the distance centre of the intersection is taken as the characteristic value. In computer, three dimensional network models are composed of large number of triangles, so the calculation of the rays and the intersecting surface is comparatively complex. In real application the demand of the calculating speed is very high, the normal method of calculating speed can hardly satisfy the demand. In this article the calculating speed is increased to maximum using co-planarity of the sample rays. In this chapter the radial method is also effectively realized using many improved concentric spheres. The last search experiment shows a difference in the original idea. It is not that the more the sample rays or the concentric spheres the better the result. Both have limits. Through many experiments the limiting value for the both is determined.The fifth chapter realized the method of spherical harmonic change. First the mathematical basis of the method is explained and analysed the reason the particular characteristic of vectors having rotational invariance in this method. Spherical harmonic change in the sphere surface can be carried out to all methods based on Chebyshev sampling. Projection of eigenvector from spatial domain to frequency domain not only enable us to achieve rotational invariance also can decrease dimension and yawp sensitivity of characteristics, thus increases the effectiveness of search. Regarding the calculating effectiveness and preciseness, the calculation of spherical harmonic change of the sphere surface is realized through the S2Kit code. In this chapter the voxel spherical harmonic change method and ray spherical harmonic change method are realized. In the comparison to grid pattern and voxel pattern can decrease the yawn effect. voxel spherical harmonic change method needs the conversion of network model into voxel model. Based on an effective method of conversion of network model into surface voxel model, this article designed a kind of calculation for converting surface voxel model into entity voxel algorithm and realized spherical harmonic change in the two types of voxel pattern separately. Ray spherical harmonic change method is an improved treatment to eigenvector derived in the chapter second. The results of the later experiments show that the application of spherical harmonic change method to entity voxel generates more effective results than to surface voxel. The application of spherical harmonic change to the radial method extremely decreased the dimensionality.The sixth chapter discusses the Derivation of framework characteristic of 3D model. Mainly, framework points are derived using the critical point derivation method from entry voxel model. First potential vector field of the voxel model is calculated and then critical point is derived using the characters of potential vector field. Lastly framework of the 3D model is obtained by connecting them. This can form connected and disconnected two frameworks. This method can be used to both the lumpish (surface and inner structure information both are known) and non-lumpish (only surface information are known) objects and clear and precise frameworks of different 3D structures can be described. At the end, an elementary description is given on the ways of getting framework characteristics descriptor using framework and their matching.The seventh chapter is the description of the visualization experiment of the feature extracting of 3D model realized in this thesis. The system is mainly divided into display module, characteristic derivation module and search experiment module. The display module can display different expression methods of 3D models, standardization process, and a part of characteristic derivation method. Through the visual effect we can easily see the real effects of the test. Characteristic derivation module can derive the characteristics of particular model database using the method described in the thesis, and can easily set into a new method. Search experiment module can execute comparison of the searches in different conditions, including the different eigenvectors, different classifying methods and different similarity measure, automatically shows evaluation index show draws Precision-Recall curve. At the end, at the base of this experiment a landscape orientation survey is carried out to the methods realized in the thesis with the best parameter derived values, and the characteristic derivation algorithm with the best effective result is applied in the 3D model search system, and effect of real search in different models are given.The thesis posses the following characteristics: -1) All the experiments carried out are to the Standard database of the Princeton University, and also compared with the results of the experiments carried out in other international institutes.2) The experiment is oriented both to 3D grid model and 3D voxel model. Though improvement and realization are carried out only few types of feature extracting method, but on the basis of this it will be easy to carry out other related feature extracting method. Thus, created a base for the further research to a deeper level in coming future in the area.