Three-dimensional Object Retrievalbased On Multiple View

Image as a multimedia medium for people to contact the information world,image retrieval is one of the important research directions.In real life,people are faced with real three-dimensional objects.Therefore,the retrieval of three-dimensional objects has gradually attracted attention,and has strong application requirements in urban planning,industrial automation,artificial intelligence and other fields.Different from the retrieval of two-dimensional images,the appearance of three-dimensional objects depends on the shape,attitude and illumination brightness of three-dimensional objects.This thesis focuses on the retrieval of three-dimensional objects and constructs a three-dimensional object retrieval based on mobile terminal system.The main research work of the thesis is as follows:(1)This paper first builds a set of three-dimensional treasure database of Shaanxi Museum.The database aims at 108 actual treasures in the Treasure Hall.There are 10800 images in the database,and a multi-view generation algorithm for three-dimensional objects is proposed.By classifying,simulating and training multiple views,the algorithm extracts the local feature description vector of an object's multi-angle image,which increases the feature and enhances the comprehensiveness of the training sample features.(2)Based on the multi-view image database of the three-dimensional object constructed in the second chapter,a three-dimensional object retrieval algorithm based on multi-view is realized.After extracting SIFT feature points,the algorithm uses TF-IDF method to generate database visual vocabulary tree index file,and constructs corresponding feature files and tree files.After that,the SIFT feature points are also extracted from the image to be retrieved,and the existing lexical tree index file is used to construct a query index to complete the search.The experimental results show that the retrieval rate of this algorithm is high.(3)An image matching point optimization algorithm based on the polar geometry constraint is studied to further optimize the matching performance under multi-view.The method utilizes the properties of the polar geometric constraint to screen the initially obtained feature matching,remove the wild spots with matching errors,and achieve the best selection of feature point pairs.Image matching point optimization is achieved by improving the feature point matching accuracy and stabilizing the number of feature points of the two images.Through a large number of experiments,the algorithm has beenoptimized for multi-view image matching points.(4)Finally,this paper builds a three-dimensional object retrieval system based on mobile terminal,and introduces the functions and principles of each component of the system in detail.The system can directly scan and retrieve real three-dimensional objects,return the name,introduction and other information of related objects,improve the practicability and convenience of the system,and have a good sense of user experience.