Multi-material Volume Segmentation for Isosurfacing Using Overlapped Label Propagation

Abstract

This paper proposes a simple and efficient method for segmenting multi-material volumes. Unlike standard image segmentation techniques, this approach aims to extract accurate isosurfaces representing the boundary surfaces between different materials. The segmented volumes obtained in this way meet the requirements of the topological correctness and geometrical accuracy for isosurfacing. The approach involves three steps: first, voxels far from the boundary surfaces are labeled; then, the labels are propagated with overlapping to ascertain which materials meet near the boundary surfaces; finally, an appropriate material is selected from among the multiple labels assigned to a voxel with adaptive thresholding. Since the method consists of iterative local operations, it is easy to parallelize for fast computation. The efficiency of the technique is demonstrated here using CT scanned objects for complex shapes.