Research On 3D Model Retrieval Method And System Based On Multi-level Feature Extraction

With the development of computer software and hardware technology,threedimensional models are more widely used in animation,mechanical,medical and other fields.With the rapid growth of the number of three-dimensional models,the retrieval and matching of three-dimensional models has become a new research direction.The existing three-dimensional model retrieval based on text description has certain limitations.The content-based 3D model retrieval technology can extract features according to the model's own structure,has intuitive description of geometric angles,and can quickly retrieve features of 3D models.It can be widely used in mechanical parts management,animation model retrieval,etc.,based on the content of the threedimensional model itself.The search method is of great significance.According to the three-dimensional model's own structure and feature extraction method,this paper studied the multi-level feature retrieval algorithm based on the content of the three-dimensional model for the three-dimensional model of triangular facets.It uses multiple machine learning methods to recognize the multi-level features and obtains different levels.Based on the recognition efficiency and accuracy of features,a threedimensional model retrieval system was developed on this basis.Low-level intermediate features The three-dimensional model original data for triangular face expressions has problems such as large differences in the number of internal constituent points and uncertain model poses.The 3D model is re-divided into grids to extract the normalized 3D model point cloud data,and the model is obtained.Low-level feature set;for large-diameter comparatively large three-dimensional models,cylindrical coordinate ray method is proposed to extract ray collections along the axis and constitute the low-level feature set of the model;for the three-dimensional model with complex surface features,multi-angle projection method is used to obtain Twodimensional view of different angles of the three-dimensional model is collected,and grayscale processing is used to obtain intermediate features of the model.Based on the middle-level feature set obtained from the multi-angle projection of intermediate-level features of the 3D model,the high-performance features of the 2D graphics are processed using the convolution depth learning network,and the intermediate features of the model are processed using the convolution deep learning network to automatically classify the 2D images,thereby achieving Pattern recognition for 3D models.Compared with other feature extraction algorithms,the dissertation method reduces the difficulty of algorithm design and improves the recognition rate of 3D models.The advanced features adopt the three-dimensional model voxelization method,transforming the three-dimensional model space into the voxel block space distribution,generating a normalized three-dimensional model voxelized data matrix,reducing the three-dimensional model feature dimension,and achieving a simplified representation of the complex features of the three-dimensional model.,Get the advanced feature collection of the model.The use of machine learning algorithms to classify and identify advanced feature sets enables accurate classification of three-dimensional models.Based on the above theoretical research,a three-dimensional model retrieval system for multi-level features is developed.The research in this paper has great practical value for image learning and pattern recognition in big data,and improves the accuracy of threedimensional model retrieval with different shapes and expressions.And efficiency has a very good theoretical significance and tool support.