Research On Rapid Filtering And Cluster Indexing Method In 3D CAD Model Retrieval

With the development of informatization in manufacturing industry,product design and manufacture based on three-dimensional(3D)model has become the main mode of discrete manufacturing.It has brought about the emergence of a large number of 3D CAD models in enterprises or Internet.Hence,how to retrieve the desired models from massive CAD model database accurately and efficiently,mine and utilize the implicite knowledge to make the reuse,is one of the crucial ways to improve the design efficiency and shorten the product development cycle.However,as a result of some undesirable properties of CAD model,such as complex structure of model,high dimensionality and massiveness of feature descriptors,the precision and efficiency of 3D CAD retrieval can not satisfy the engineering application till now.In order to promote the granularity of feature description,the discriminative power for models' local parts and the searching efficiency in 3D CAD model retrieval,this thesis carries out a systematical research on high efficienct retrieval of 3D CAD model,mainly including three key aspects—feature extraction and similarity measure,clustering division of model database space and indexing technique for model database.The structure of this thesis is arranged as follows:1)An extraction method of hierarchical feature descriptor for 3D CAD models is proposed.Initially,a definition called Labelled Attribute Adjacency Graph(LAAG)is presented,based on which 3D CAD models are transformed to LAAGs by extracting the essential information from B-Rep.Next,the models in training dataset are segmented into different regions according to their corresponding LAAGs with an improved segmentation method,which are represented as local feature vectors using graph spectrum.Finally,the hierarchical feature descriptor(HFD)of each library model is constructed,which is composed of CAD model Spatial Bags-of-Words(CMSBoWs)as top-level descriptors and local graph spectrum as bottom-level descriptors.A two-level searching mechanism for CAD model retrieval is employed to validate the effectiveness of HFD.Experimental results show that the proposed HFD has a good discriminative power for models' local parts,moreover,it facilitates the similarity measure between models.Therefore,this method lays a good foundation for promoting the efficiency and precision of 3D CAD model retrieval.2)A clustering and retrieval method inspired by fish swarm for 3D CAD models is proposed.In order to tackle the inefficiency problem of CAD model retrieval,which is leaded to by linear search and similarity measure,this method partitions model database into a number of sub spaces with clustering algorithm.Based on the extracted HFD above,the mathematical model for partitioning CAD model database is defined.Then,an improved Fish Swarm inspired optimization method,based on the global bulletin board guiding along with the fuzzy c means modifying,is presented to partition the whole database into sub model bases.In the clustering based model retrieval,the query model is firstly classified into the attached sub model bases with membership function,and then,the results which are the most similar to the query are retrieved in these limited spaces.From the experimental results,it can be seen that the proposed methods of clustering is good at dividing the searching space of CAD model database.The clustering based retrieval methods promote efficiency obviously;moreover,its precision is almost as good as the linear search approach.Furthermore,the proposed method can be also useful to other aspects of model data management such as model navigation,browsing,knowledge discovery,etc.3)A clustering hierarchical decomposing based 3D CAD model database indexing(CHDC-Tree)is proposed,based on which an efficient pruning and search approach is also presented.Because the clusters tend to overlap each other especially for high-dementional data,the promotion for efficiency of CAD model retrieval is limited,if it is singly based on clutering algorithm;moreover,the parameters of clustering also affect the efficiency greatly.Hence a clustering hierarchical splitting and indexing approach is proposed,through which the model database space is partitioned more finely to reduce the number of model access.Initially,the clusterings are further splitted into a number of sub spaces,each of which has roughly the same number of models,and meanwhile,the index called CHDC-Tree is constructed,according to the distances distribution among feature descriptors.Therefore,CHDC-Tree is a balanced tree and can tackle the curse of dimensionality.Then,in the CHDC-Tree based model retrieval,an auto search and pruning method based on depth-first search along with dynamical revising query radius is firstly employed to retrieve the candidate models,and the remaining candidates are verified using optimal matching algorithm according to their local features.Experimental results show that the proposed method has good performance on indexing and pruning for 3D CAD model retrieval;moreover,it also has a good robustness to the parameters of clustering and indexing algorithms.For massive CAD models retrieval,the proposed method has significantly better efficiency than the linear search method,so it can support the effective reuse of CAD model in engineerings.4)A graph indexing and filtering mechanism based partial retrieval method for 3D CAD models is proposed.Initially,3D CAD models are transformed to LAAGs from B-Rep.Then,based on graph spectrum approach,LAAGs of each CAD model and their vertices' local structures are encoded to feature vectors respectively,which are suitable for indexing or filtering of partial retrieval.Lemmas about partial matching between CAD models are proved,upon LAAG and its vertices codes respectively.Then,the filtering rule based on LAAG and its vertices codes is also presented respectively,moreover,an index tree called GraCode-Tree is constructed,according to the LAAG based filtering rules.Finally,in order to enhance the efficiency of partial retrieval,a 2filters-verification mechanism is presented.Experimental results show that the precision of this method is good;moreover,it speeds up partial retrieval for CAD model significantly.