Constructing Smooth Non-Manifold Meshes of Multi-Labeled Volumetric Datasets

This paper presents a method for constructing consistent non-manifold meshes of multi-labeled volumetric datasets. This approach is different to traditional surface reconstruction algorithms which often only support extracting 2-manifold surfaces based on a binary voxel classification. However, in some – especially medical – applications, multi-labeled datasets, where up to eight differently labeled voxels can be adjacent, are subject to visualization resulting in non-manifold meshes. In addition to an efficient surface reconstruction method, a constrained geometric filter is developed which can be applied to these non-manifold meshes without producing ridges at mesh junctions.