Research On Some Key Technologies Of Shape-Based 3D Model Retrieval

Improved modeling tools and scanning devices are making the acquisition of 3Dmodels easier and lower expensive. 3D graphics hardware and CPUs have becomefast and cheap enough that 3D data can be processed and displayed more quickly onPCs. It leads to high demand for 3D models from a wide range of sources.Under the background of Internet, 3D model retrieval on the base of intrinsicproperties is proposed in applications of 3D model. It is new extension ofcontend-based image retrieval.In the research of 3D model retrieval, the key problem is to get invariance ofgeometric transformations of rigid bodies, at the same time the feature vector spaceshould be easy to be extended, which will facilitate the massive model retrieval. Dueto the complexity of 3D objects, the acquired features should contain adequateinformation.This dissertation researches mainly in three aspects, describing the shape of 3Dmodel;getting the shape properties of 3D models;and indexing 3D models by usingthe description of its shape properties.Normally, four steps are included in a 3D models retrieval system, i.e. posenormalization, feature extraction, similarity comparison and user interface. Thisdissertation focuses on the extracting method of global geometric features of 3Dmodels and its applications in 3D model retrieval. The main contributions of thisdissertation are summarized as follows:A novel 3D orthogonal polynomials moment, called 3D U system moment isdefined, and is used for 3D model retrieval. As a kind of complete orthogonalpolynomials system, U system has many good properties. According to U systemtheory, the fast calculation of U1 system moments is established.A novel feature vector-extended distance is proposed. It is obtained by forming aspherical function of maximum distance on the surface of the model. Then, sphericalharmonic transform is used on the sphere. The energy of the coefficients of sphericalharmonics forms the feature vector.Electric force feature is proposed, which is based on electric field theory.Supposing 3D model as charged body, we get the electric field force distribution byplacing some testing charges around the 3D model. The force distribution is thefeature of the 3D model. This representation has several benefits. First, it is invariantto scale and rigid transform by calculating the spherical harmonic transform. Second,it can represent complex and ill-defined models because of its physical background.3D orthogonal spline moments are developed as new shape descriptors for 3Dmodels. They are robust to translation, rotation and scale, because they areconstructed in the general 3D rotation invariant moments frame. Experiments showthat 3D orthogonal spline moments have higher discriminating power compared with3D wavelets moments.A novel preprocessing method of a 3D model retrieval system is created. The 2Dsketches are used as three-view drawings of a 3D model. And on the basis of sampleof three-view drawings, the 3D voxel model is reconstructed. It can avoid themulti-comparisons of 2D sketches. Furthermore this technique can be a solution ofinput interface of 3D model retrieval.