Research On Semantic Annotation Methods Of The Three-dimensional Model

A sufficient condition is provided for people to handle and share data by the increasinglyadvanced internet technologies. Faced with a huge three-dimensional model database, toachieve optimal management of huge amounts of data, to help users quickly and accuratelyobtain a three-dimensional model with the design intent, to achieve resource-reuse, becomeurgent problems to be solved. In order to get the desired retrieval effect, semantic-based3Dmodel retrieval attracted wide attention of researchers. This search method is mainly based onsemantic representation of objects to retrieve and match the corresponding model data, andobject-based semantic representation must rely on the three-dimensional model of semanticannotation technology.Considering the current shortcomings of various three-dimensional models’ overallsemantic annotation methods, two automatic semantic annotation methods have beendesigned in this article. They are3D model semantic annotation method fusing shape andsemantic relevance, and3D model semantic annotation method on both skeleton structure andcorrelation transitivity. The second method has been improved and innovationed on the basisof the first method, and has higher accuracy and efficiency compared with the first method.During the model matching process of the first3D’ semantic annotation method,topological features and shape features are used to match the similar models. In the semanticannotation process, the relevant base concepts and algorithm have been proposed. Therelevant base algorithm is used as one of standards that measure the degree of correlationbetween the unlabeled model and semantic vocabulary. Labeling process as follows: Firstly,those three dimensional models that are similar should be picked out by usingthree-dimensional models’ topological features and shape features. Secondly, to calculatequite similar vocabularies in the vocabulary sample, and extract the vocabularies which arehigh degree of similarity to the unlabeled model, then to form the first set of vocabularies.Thirdly, the model-semantic vocabulary related base is calculated based on the semanticcorrelation and heir principle among the three-dimensional models, furthermore, vocabularywords which are in a high degree of correlation are determined in the sample, then to form thesecond set of vocabularies. The last, semantic annotation of the unlabeled model isdetermined by both of the two sets of vocabularies. The final experiment showed this methodcan enable semantic annotation of the unlabeled model more accurate.The second method uses the topological features and geometric features of the model’skeleton structure to complete matching process. During the model matching process of thefirst3D’ semantic annotation method, the algorithms for the calculation of semantic similaritybetween words are proposed based on the semantic correlation transitivity principle among the three-dimensional models, which combines the probability of similar words,models-semantic direct correlation and overall similarity of the models. Labeling process asfollows: Firstly, those three dimensional models in the sample that are similar to the unlabeledmodel should be picked out by topological properties and geometric properties. Secondly, tocalculate quite similar vocabularies in the vocabulary sample, and extract the vocabularieswhich are high degree of similarity to the unlabeled model by the algorithms for thecalculation of semantic similarity between words. The algorithms combine the probability ofsimilar words, models-semantic direct correlation and overall similarity of the models.Thirdly, semantic annotation of the unlabeled model is determined by extracting vocabularieswith a high degree of similarity to the unlabeled model. A series of experimental resultsvertify the accuracy and the robustness of the method.