Incomplete Model Retrieval Based On Hole Filling

As 3D models are widely distributed in the real world,it is easy to appear that the data of the 3D model stored in the database is incomplete due to external force damage,natural damage and insufficient scanning.How to use the incomplete 3D model to detect the corresponding complete 3D model in the 3D model database is a difficult problem at present.Therefore,this paper takes the incomplete 3D model as the research object and proposes the following two retrieval methods:(1)A 3D incomplete model retrieval method based on hole filling and clustering is proposed.The method firstly detects and fills the holes of the incomplete 3D model,preprocesses the 3D model after filling,and directly calculates the curvature of each point,which is used to describe the characteristics of the filled 3D model.Then,K-means++ clustering algorithm is used to cluster the retrieval model and the target model,and a calculation method of C_DIS similarity measure is proposed to calculate the similarity between the models by class.In this method,the filling holes can supplement the incomplete model,and the implicit surface obtained by solving the radial basis function can guarantee that the filling is as close to the complete model as possible.The curvature has geometric invariance.Clustering not only avoids the same curvature when the points match,but also guarantees the one-to-one matching between the retrieval model and the target model.C_DIS similarity measure is proposed based on the calculation principle of Hausdorff distance.The model is regarded as a set of several sets of points,with high similarity in class and low similarity between classes,and more accuracy in inter-class matching.(2)A 3D incomplete model retrieval method based on hole filling and reliability constraint is proposed.In this method,the holes with complex structure were firstly divided into sub-holes with simple structure,and then the model was preprocessed after filling.Then the bounding box is built for filling the 3D model and the data is simplified by dividing it into several small cubes.Since the data points before the simplification are not all owned by the 3d model itself,the concept of reliability is introduced to measure the reliability of the simplification points and calculate the similarity between the models and the reliability of the final retrieval results under the constraint of reliability.In this method,large holes and complicated structures which cannot be filled directly are segmented according to their local details,and sub-holes are filled respectively to ensure high accuracy.Data simplification improves computing speed.The introduction of the concept of reliability quantifies the results of abstract similarity,which is helpful to measure the reliability of the retrieval results.Both of the two retrieval algorithms established the filling 3d model as an intermediate link between the incomplete 3d model and the corresponding complete 3d model,which built a bridge between the incomplete 3d model not stored in the database and the complete 3d model stored in the database.The effectiveness of the proposed algorithm is proved by experimental evaluation of the two retrieval algorithms.