Patch Based Collaborative Normal Filtering For Feature-Preserving Mesh Denoising

Mesh denoising is a traditional problem in digital design and manufacturing but still not wellsolved.With the development of various scanning devices,it is more convenient for people to obtain three-dimensional data.However,due to the limitations of the scanning device itself,the influence of external environment and the errors of various surface reconstruction algorithms,the obtained threedimensional model often has noise.Before using these mesh models for subsequent geometry processing,we need to denoise the corrupted surface.The main challenge of mesh denoising is to remove noise while minimizing the intrinsic properties of the surface,such as sharp features.However,both mesh noise and sharp feature are high frequency information,and it is generally difficult to distinguish them well.In order to solve the problem,this paper propose a novel patch based normal collaborative filtering algorithm for mesh denoising.It is inspired by the geometric statistics which show that there are many similar patches exist on underlying surface of the noisy mesh.This algorithm can be divided into two steps: first,we find similar patches of the mesh model globally and make their normal invariant to rotation transformation,input the normal vector information of these patches into the matrix,and perform low rank recovery operation;then,Then,a new mesh patch guide normal vector is calculated according to the restored matrix,and the normal mesh domain is filtered by the normal.Finally,the position coordinates of the mesh vertex are updated in terms of the filtered normal vector.The key of the algorithm is how to extend the low rank matrix recovery in the traditional two-dimensional image to the three-dimensional mesh processing,and how to effectively recover the original matrix from the ultra-low-dimensional noise matrix.Finally,we carry out a large number of experiments on the synthetic model and the real scan model,and clarify the effectiveness of the algorithm from the perspective of visual effects and quantitative analysis.