Research On 3D Shape Description Based On Anisotropic LBO And Its Application

With the rapid development of Internet technology and 3D scanning technology,3D models become one of the new-generation multimedia data types.3D databases are common and easier to acquire in different fields,such as computer-aided design,molecular biology(3D protein models),computer graphics,medicine and archeology.Therefore,3D model retrieval,matching,recognition and classification become top issues in current research.At present,in the face of great database of 3D models,people do not only focus on how to design 3D models,but also pay more attention to accurately describe 3D models and quickly search the target model in the massive model library,which is also the focus of this article.In recent years,researchers have proposed many methods for the description and retrieval of non-rigid 3D models,but these methods still have some limitations.On the one hand,the existing methods are based on isotropic Laplace-Beltrami operator(LBO)and cannot fully characterize the intrinsic properties of non-rigid models.On the other hand,most methods are difficult to achieve the accuracy of model description and the efficiency of retrieval.In order to solve the above problems,this thesis introduces an anisotropic Laplace-Beltrami operator.It is based on the isotropic LBO.While keeping useful properties of the standard Laplace-Beltrami operator,it introduces variability in the directions of principal curvature,this operator improves the accuracy of acquiring the features of the 3D model,opens the possibility to effectively replace the omnipresent Laplace-Beltrami operator in many shape analysis methods.In addition,this thesis also improved the discretization method of anisotropic LBO,and experiment with 3D data in the SHREC'14 database(International Public Database).Further,we demonstrate the effectiveness of our new operator.In addition,this thesis proposes an Combined point Signature based on Independent component analysis(ICS)and Shape Descriptor based on Independent component analysis(ISD).Based on Wave Kernel Signature(WKS)and Scale Invariant Heat Kernel signature(siHKS),this thesis constructs WKS and siHKS based on anisotropic LBO,and combines them to form a new point signature.At the sametime,we extract the independent components of the combined point signature by Independent Component Analysis(ICA)to ICS and ISD.The ICS point signature and the ISD shape descriptor can capture both the global information and the local information of 3D model.Therefore,they can accurately describe the non-rigid 3D shape,and have the performances of high efficiency and simplicity.Finally,generated ICS point signature and ISD shape descriptor are applied to the 3D model.We use the 3D model in the SHREC'14 database as experimental object,and the ISD shape descriptor are combined with Large Margin Nearest Neighbor to achieve classification and retrieval of 3D shapes.They turn out that the effectiveness of the ISD shape descriptor,and it also demonstrate the effectiveness of the anisotropic LBO in extracting and describing 3D features.The ICS point signature and ISD shape descriptor reduce the complexity and difficulty of computing high-dimensional data,and they can quickly and accurately describe the features of non-rigid 3D shapes,and improve the 3D retrieval accuracy.