A lightweight approach to repairing digitized polygon meshes

When designing novel algorithms for geometric processing and analysis, researchers often assume that the input conforms to several requirements. On the other hand, polygon meshes obtained from acquisition of real-world objects typically exhibit several defects, and thus are not appropriate for a widespread exploitation. In this paper, an algorithm is presented that strives to convert a low-quality digitized polygon mesh to a single manifold and watertight triangle mesh without degenerate or intersecting elements. Differently from most existing approaches that globally resample the model to produce a fixed version, the algorithm presented here attempts to modify the input mesh only locally within the neighborhood of undesired configurations. After having converted the input to a single combinatorial manifold, the algorithm proceeds iteratively by removing growing neighborhoods of undesired elements and by patching the resulting surface gaps until all the "defects" are removed. Though this heuristic approach is not guaranteed to converge, it was tested on more than 400 low-quality models and always succeeded. Furthermore, with respect to similar existing algorithms, it proved to be computationally efficient and produced more accurate results while using fewer triangles.