3D Shape Matching And Retrieval Based On Point Bidirectional Features

The three-dimensional (3-D) information technology, which is known as the fourthgeneration of multimedia technology, includes3-D digital models and3-D scene technology.It is gradually becoming the main stream of media technology and has an important influenceon human life, work and entertainment. With the popularization and application of computernetwork technology, the3-D modeling technology has become increasingly mature, and thenumber of3-D models in the virtual world is expanding with geometric growth. How tosearch the required3-D model from large-scale model library or the Internet has become a hotbut difficult spot in the current computer graphics communities. This thesis performsintensive research on feature extraction, shape matching, similarity measure in3-D modelretrieval process, which has important theoretical significance and practical value.Firstly, the classic3-D model feature extraction algorithms are introduced in detail in thisthesis. These features are distinctive and have applied in different fields successfully. Then,the concept of bidirectional feature is introduced, and the detail construction process of theproposed Point Bidirectional Features (PBFs) and their representation are described in details.Secondly, this thesis presents the application of the proposed PBFs on3-D modelmatching field. Comparing with the classic improved ICP (Iterative Closest Point) registrationalgorithm and3-D-shape-content based model matching algorithm, the proposed methodshowed better competitiveness for matching non-rigid deformed models.Finally, this thesis designs a novel PBFs-based3D model retrieval algorithm. The firststep of this method is to extract the PBFs for the key points. Then the KNN(K-Nearest-Neighbor) search strategy is adapted to points matching in the feature space. The models similarity measurement is also designed to search for the query model. Test this3-Dmodel retrieval algorithm on Princeton Segmentation Benchmark (PBS)3-D database, McGill3-D database and SHREC073-D database, respectively. Introduce a variety of evaluationcriteria to evaluate the retrieval results and compare with some other3-D shape retrievalmethods. The experiment proved that the proposed3-D shape retrieval algorithm had betterretrieval efficiency.