Mesh Stitching Preserving Sparse Features

With the rapid development of the3D industry, there is a rising need for fast and convenient methods to construct3D models. Traditionally, tools to create3D models are either confined to artists or based on stereo vision techniques which are not so practical. While the increasing use of3D scanners may alleviate the contradiction, they are still unable to be widespread because of the high costs. As a consequence, shape modeling by part composition is gaining more and more attention.In Digital Geometry Processing, there have been many advanced works about shape composition. While existing approaches have mostly focused on smoothly blend different parts, they cannot always guarantee a perfect output, especially when the original parts have sharp features.In this paper, we introduce a feature-preserving method for the composition of piecewise smooth meshes. We assume that the meshes involved have been manually brought into the proper positions by the user, and they have no overlap. To preserve the shape, we divide the composition into two stages:smoothly stitching and feature restoring. In the smoothly stitching stage, we generate an additional mesh connecting the two different parts using conforming Delaunay triangulation and Laplacian interpolation. In the feature restoring stage, we first recover the feature in the normal field using L1-sparse, and then deform the connected mesh to match the new normals using bilateral filtering. Before outputting the result, we will also remesh the connected mesh to improve the sampling and connectivity quality.The main contribution of this paper is to propose a new question about feature-preserving mesh stitching and this topic may deserve our long-term research.