Research On The Description And Retrieval Methods Of Non-rigid 3D Data

With the rapid development of computer graphics and related software and hardware technologies,it becomes easier to acquire 3D models.The importance of 3D models and market demand is increasing,3D data quickly becomes the fourth generation of multimedia data after images,sounds and videos,which is widely used in image processing,3D online games,CAD,e-commerce,industrial manufacturing,virtual reality and many other fields.Surface description,analysis,matching,and retrieval of 3D models have become research focus.Now people are no longer limited to the study of how to build 3D models,they pay more attention to how to reuse and share existing model resources for application research,the core of this is to describe and retrieve the target model in a large number of databases.This paper mainly studies the description and retrieval methods of non-rigid 3D models based on Laplace-Beltrami operators.We introduce the existing description and retrieval methods of the non-rigid 3D model.Among non-rigid 3D model description methods,the siHKS and WKS description methods based on Laplace-Beltrami operator are better methods.siHKS can describe the overall shape information of non-rigid shape,and maintain good stability for noise,topological disturbance and partial missing.WKS focuses on describing local feature information and pay more attention to the description of the details of the 3D model.This paper bases on the Laplace-Beltrami operator and uses the WKS and siHKS point signature to obtain the combined point signature.Then analyze the combined point signature to generate PCS point signature by Principal Components Analysis.We use SHREC'14 human model database as a detailed example to elaborate PCS point signature and the experimental verification of PCS point signature is more effective.The 3D model mesh vertex will generate a large number of point signature,so the 3D model description and retrieval calculation are large and inefficient.Based on the PCS point signature,this paper improves the method of weighting the shape descriptor by area,we calculate the curvature of the vertex of 3D model and employ shape curvature as the element of weight in the construction of PSD shape descriptor.Generated PCS point signature and PSD shape descriptor are applied to the 3D model retrieval.We describe 3D model retrieval process,search technology and retrieval performance evaluation index.We use Large Margin Nearest Neighbor(LMNN)approach of the resulting shape descriptors in SHREC'14-Shape Retrieval of Non-Rigid 3D Human Models.Selecting PCS point signature can acquire better comprehensive and accurate shape features of the 3D model,reduce the dimension and calculation amount.The results of 3D model retrieval based onPCS point signature and PSD shape descriptor show that the PCS point signature and PSD shape descriptor can effectively extract the shape feature information of non-rigid 3D model and improve the accuracy of retrieval.