Research On 3D Model Retrieval Based On Sparse Representation

Improved modeling tools and scanning devices are making the acquisition of 3D models easier and lower expense. It results in high demand for 3D models from a wide range of sources.3D model reuse is an effective way to obtain a new model. How to retrieve the needed 3D models from the databases is growing up to be an important issue. This thesis focus on three aspects that are model smoothing preprocessing, feature extraction and similarity measurement of the general 3D models in mesh representation. The major contributions are described as follows.1. Extend the technique of sparse representation for 3D Terracotta Army model smoothing preprocessing which uses Laplacian and wavelet base to construct a sparse dictionary. It establishes the overall framework of 3D model processing based on sparse representation. Coherent parameter value and the comparative experiment results illustrate the effectiveness of this sparse dictionary in theory and practice. This method can get the better smoothing effect than spectral method using fewer reconstruction coefficients.2. Propose a 3D model smoothing algorithm based on Laplacian coordinate. Feature point labeling process is formulated as an optimization problem when setting the normal mean curvature of 3D model’s vertex is zeros and using (?)1 norm minimum constraint. The new weight function and energy function are constructed for 3D model smoothing. Experiment and a variety of evaluation results show that the algorithm can better to maintain geometry details of 3D Terracotta Army models.3. Present an effective feature extraction algorithm based on multi-features fusion for 3D model retrieval. Global feature is extracted using spherical coordinates and rays method. Local feature is extracted using local radial distance mapping a gray level image. Global feature and local feature are fused into a new shape descriptor by kernel function. The innovative shape descriptor includes all information of the original descriptor which can describe 3D model more comprehensive. Experimental results show the algorithm has better retrieval effect.4. Propose a sparse matching algorithm for 3D model retrieval system. The method transforms the similarity matching process of feature vector into get the solution of underdetermined equation system. Utilizing the energy function minimum and setting slack variable, the problem changes into a second order cone programming problem. Optimum solution is the ultimate similarity matching result. Using sparse processing for feature database and feature vector can narrow the time complexity of the algorithm. It can speed up while feature matrix partitioning based on the category information. Experimental results show that the algorithm has higher precision ratio and more robust.This work is supported by the key project of the National 863 project fund. The related algorithms are verified in the processing of 3D Terracotta Army models.