A Research On Content-based 3D Model Retrieval And Clustering

With the rapid development of network information technology and computer software and hardware technology,the number of 3D models and the demand for model retrieval are also growing rapidly.The search engines commonly used today mainly provide retrieval services based on semantics,but retrieving 3D models based on text descriptions has certain limitations.Such retrieval methods are too subjective.Therefore,content-based 3D model retrieval methods have emerged at the historic moment.The research is launched from three aspects: model retrieval,image set of 3D models extraction,and model library clustering.The main contents and innovations are as follows.(1)A 3D models retrieval method based on curvature distribution from global aspects is proposed.First,according to the distribution of the global discrete curvature,the vertices of the model are divided into four categories.Then build a geodesic distance distribution histogram for each type of vertex.Finally,calculate the model similarity with weight.Experiments show that this method is superior to common vertex segmentation methods and has better retrieval effect on smooth and flat models.(2)A 3D models retrieval method based on the salient points from local aspects is presented.First,the salient points of the model are extracted based on the geodesic distance.Then the local area is divided according to the salient points,the distance-angle feature matrix is constructed in the local area.Finally,the similarity of the model is calculated using the earth mover's distance(EMD).Experiments show that the algorithm can effectively extract local area to characterize the entire model,and has high retrieval accuracy than similar methods.(3)A framework for extracting image set of 3D models based on web page is proposed.Use HTML5 and ThreeJS to browse the model anytime and anywhere,and quickly extract the model image set.(4)A clustering method of 3D model library based on multiple features is studied.First,take multiple model images from multiple viewpoints.Then three image features and two model features are extracted.Finally,the model library is clustered using the subspace clustering algorithm of arctangent approximate rank.Experiments show that this method basically meets the preprocessing requirements of the 3D model library and is superior to the traditional model clustering method.