Research On 3D Model Retrieval Algorithm Based On Visual And Topological Features

With the rapid development of virtual reality technology and Internet technology,the number of three-dimensional(3D)models and the demand for retrieval are increasing rapidly.The 3D model has gradually become a new type of multimedia data.At the same time,a huge amount of shared 3D models appear on the Internet.The common search engine does not provide efficient 3D model retrieval service.The 3D model retrieval technique based on content is widely concerned and becomes an active research field.Therefore,how to quickly and effectively find out the 3D model according to user’s requirements from a large number of 3D models and to realize the reuse of resources have become an urgent problem,and it has important theoretical and realistic significance.In order to achieve the goal,we propose two algorithms for 3D model retrieval:(1)In this paper,a new method for 3D model retrieval is proposed by combining the projective invariant of the point feature and spatial statistic distribution characteristic.The topological feature and statistic characteristic of the model are fused into a unified framework.Firstly,the posture and scale standardization of each model are completed during the process of the model data pre-process.Then,the spatial division and random sampling method is introduced to build the topological structure projective invariant of the point feature of 3D models.It improves the accuracy and robustness of 3D model retrieval.The Recall-Precision curve of our algorithm and other algorithms are analyzed.It proves the effectiveness of our algorithm.(2)We propose an approach to three-dimensional building model retrieval based on topology structure and view feature.This study uses two filter steps to finish the building model retrieval.Firstly,the distance transformation method is used to extract the topological structure of 3D building models.The obtained skeleton points of 3D building models are classified by our homocentric sphere method.3D building models are filtered for the first time based on the distribution features.Secondly,3D building models are projected onto two-dimensional(2D)images from the viewpoints of skeleton points.SIFT algorithm is used to extract feature points from 2D projection images,and the idea of BOF method is used to compare the projection images.3D building models are filtered for the second time.Finally,we can get the retrieval result(the matched models).And the experiment result shows the effectiveness of our method.