Research On Graph Matching Based 3D Model Retrieval

With the rapid development of computer technology in recent years,the development of graphics hardware and network has derived the extensive application of 3D model in different fields,and also makes the number of 3D data in the local data storage and Internet online data storage to provide a rapid increase in the number of.Due to the large number of three-dimensional objects,it leads to an urgent need for efficient three-dimensional model retrieval and recognition techniques.The study of 3D modeling has been going on for decades,and the methods of 3D model retrieval and recognition can be divided into two paradigms: model-based and view-based.The early approach is based on a model-based algorithm,which requires the condition that the 3D model is clearly known.Another three-dimensional model retrieval method is view-based,where each three-dimensional object is represented by a set of multi-view views and extracts features from it.Compared with the model-based 3D model retrieval,the view-based 3D model retrieval is more flexible than the non-mandatory of its virtual three-dimensional model information.On the basis of the research at home and abroad,the research on 3D-based model retrieval algorithm based on view still has great challenge.In this paper,we focus on the three-dimensional model retrieval algorithm based on multi-graph matching.Firstly,the multi-view view of 3D model is taken as the input,and the characteristics of convolution neural network are extracted from all the views of 3D objects,and the visual characteristics of the database are described.Based on the principle of similarity maximum,the principle of unity or binary matching consistency,the initial matching matrix is used to calculate the best one or two binary stepwise regularization to find the best intermediate level graph in the multi-view.Again,according to the node matching consistency,The mask,the best intermediate level map,the conversion of the unity of matching or binary matching consistency,repeat the above steps to build similarity hypergraph or consistency hypergraph,using the maximum spanning tree algorithm to generate the final matching matrix Finally,the final matching matrix is normalized,and the obtained sequence is the similarity between the two three-dimensional models in the database,and the similarity degree is sorted to realize the retrieval of the three-dimensional model.This method effectively avoids the generation of deformation noise and outliers,and combines the graph matching with the whole information of the 3D model database to improve the efficiency and precision of graph matching retrieval.Extensive comparison experiments on ETH,NTU and MVRED dataset can support the superiority of the proposed method.