Technology Of Content-based 3D Model Retreival

With the use of 3D modeling software and the development of 3D scanning technology, it is more and more easier to get 3D models. A lot of 3D models are produced and used in various fields. 3Dmodels have become a new multimedia data type following picture, sound and video. We must resolve the problem of how to find a model that we need from so many models, and then the technology of 3D model retrieval was born. At present, the focus of 3D model retrieval is the method based on content, and the content-based methods could divide into many sorts. A whole retrieval process include preconditioning, extracting feature vectors and matching similarity. In this paper we research and discuss some sorts of content-based methods, and propose several new methods.Firstly, this paper expatiates the background and meaning of 3D models retrieval. Secondly, we introduce key technologies in every step of model retrevival. The preconditioning of 3D models includes translation transformation, rotation transformation and scaling transformation. There are many sorts of methods that extract feature vectors, and there isn't a uniform mode to classify them. In this paper we divide them into 7 kinds and expatiate every kind. The last step is matching similarity and the main methods of computing are Manhattan distance, Euclidean distance and Hausdorff distance.Between the technologies of extracting feature vectors, the method based on spherical harmonic transform is very important. When we use this method to extracting feature descriptors, it is unnecessary to rotate the model. Kazhdan's algorithm voxelized the model firstly, and then transform the result into sampling values of spherical harmonic transform. In this process it will engender error. This paper spherical voxelized in the shpere surrounding the 3D model and the errors generated in converting from rectangular coordinate to spherical coordinate can be avoided.The skeletonization algorithm of 3D model is based on topology structure. The key of this method is extracting the skeleton of models. There are two common methods: distance transform and Reeb graph. In extracting the skeleton of 2D images distance transform has a good effect, but the time complexity is exorbitant when we apply it in 3D models. In this paper we propose an approximate method of distance transform to extract skeleton points of 3D models and treat them as feature vectors to retrieve models.In nature and our lives, many objects have symmetries. Symmetry is important character of object and important portion of human vision perceptual organization system. The descriptor based on symmetry can commendably reflect shape information and be used in 3D model retrieval. This paper use two methods to extract symmetry planes of 3D models. The position information and normal vectors are regarded as feature vectors. The experiment indicate that this algorithm could retrieval 3D models effectively.