Research On Graph Matching-based 3D Model Retrieval

With the advances in 3D rendering and visualization technologies,3D models have been widely applied in computer-aided design,entertainment industry,mechanical design and so on.There are trillions of 3D models on the Internet.How to achieve accurate,efficient and fast retrieval of the required model in the massive 3D model dataset has become an urgent problem.At present,a variety of effective algorithms have been proposed in the field of 3D model retrieval,which can be classified into two categories: model-based method and view-based method.In particular,view-based 3D model retrieval methods has been widely studied because it is very flexible and easy to implement using multiple views for model representation.In this paper,we first investigate several popular 3D model retrieval systems and 3D model retrieval international evaluations,and then summarize the general framework of view-based 3D model retrieval.Secondly,we propose two effective 3D model retrieval algorithms based on graph matching: 1)Multi-Clique Graph(MCG)Matching for 3D model Retrieval: In the first step,the 3D model is calibrated,positioned and dimensioned,and then the virtual camera array is used to obtain the 2D view set of the 3D model from different perspectives according to the orthogonal projection method,and the Zernike moment feature of the view is extracted.Each view of a 3D model is represented as each node of the original graph,clique discovery can be realized by utilizing a hierarchical agglomerative clustering(HAC)method,and then MCG Matching can be converted to optimize the Integral Quadratic Programming problem(IQP)with affinity constraints,and we regard the result of the matching as the similarity between the two different 3D models.2)Hyper-Clique Graph(HCG)Matching for 3D model Retrieval: MCG can be considered the second order HCG,we straightforwardly extends the first-& second-order clique graph structures to the general and arbitrary high-order clique-graph representation and matching,in which each clique can be considered as one hyper-node that contains a set of neighboring nodes and each hyper-edge can directly link multiple cliques.Specifically,we embed the clique relations of arbitrary orders in a high-order similarity tensor in a recursive manner.Then,we formulate the objective function of HCG matching with respect to two latent variables: the latent clique structure information in the original graph and the similarity measure of clique sets from pairwise HCGs.In addition,we suitably adopt the affinity-preserving reweighted random walks to optimize the objective function.Finally,the matching result of HCG is obtained and used as the similarity between the two 3D models.In this paper,we selects the more popular 3D database to evaluating our method,and compare with some classical algorithm.A lot of experimental results reveal that the advanced algorithm is superior and effective.