User-guided feature sensitive hole filling for 3D meshes

Abstract

The 3D models obtained from the 3D scanners are often noisy and have flaws such as gaps, holes, self-intersecting polygons, etc. These defects make them undesirable for various applications. Thus many post processing techniques are needed to apply to improve the quality of the 3D models so that they can be usable in 3D applications. Hole-filling is an important task among them. Most hole filling approaches use techniques to fill up a hole and to smooth the surface without recovering fine features of the original 3D model. This thesis proposes a methodology which can fill holes in a 3D mesh and at the same time preserve sharp features of the geometry of the original model. The main idea is that we reconstruct possible feature curves in the missing parts of the given mesh before filling the hole with smoothing surface. The feature curves in the missing part are reconstructed by extending salient features of the existing parts. Then the hole is partitioned into several smaller and more planar sub-holes divided by the feature curves. By this way, the sub-holes make the hole filling process be more efficient and accurate. User intervention is needed to design fine features or to guide feature curve reconstruction wherever ambiguity exists or results are unsatisfactory. The ambiguity comes from multiple choices to build feature curves out of crest lines of the existing 3D mesh. Our hole filling techniques is different from other existing techniques as features are taken as the first subject to reconstruct, which eventually drive the feature-definite surface filling process.