Research On KAZE Feature Of 3D Mesh

With the rapid development of computer hardware and software technology,the 3D model not only has been widely used in the computer aided design,life sciences,medicine,chemical,industry,military and other fields,but also can be seen everywhere in our life,such as 3D game animation,virtual reality and so on.Compared with the 2D images,3D model and 3D scene can provide more and richer visual perception details.At present,the technology about 3D model develops rapidly,3D model retrieval technology involves artificial intelligence,computer vision,pattern recognition,and many other fields,and it has became a hot topic again after the image and video retrieval.The key issue of 3D model technology is feature extraction method,an efficient and perfect 3D model feature can greatly improve the efficiency.This issue do the research on the 3D model feature extraction method and 2D image features,we propose a new 3D model feature extraction approach,called Mesh-KAZE features.In two-dimensional space,for the image feature,SIFT feature has good scale invariance,Mesh-SIFT has extended this approach to 3D mesh successfully.Similar to the SIFT algorithm,the scale space is constructed by Gaussian filter,and the Gaussian filter for meshes is approximated as subsequent convolutions of the mesh with a binomial filter.KAZE is a feature detection algorithm that is more stable than SIFT feature,its nonlinear scale space is constructed by anisotropic diffusion and Additive Operator Splitting(AOS)techniques,it is more stable than linear space.The goal of this paper is expand the KAZE features from 2D to 3D space,which presents a new multi-scale feature extraction method on 3D mesh in nonlinear scale space,we adopt a non-iterative feature preserving filtering technique based on robust statistics for 3D mesh smoothing,so that we construct the nonlinear scale space on mesh.The scale space is more scale invariance.We extract the model local features,it can enhance the local expression for the 3D model,and we can get descriptor whose scale invariance is better,the descriptor can improve the descriptive skills for the 3D model,We can obtain a more robust 3D model feature descriptor with stronger discrimination ability.We develop a 3D model retrieval system with this feature,which improve the retrieval efficiency and accuracy.