Hierarchical Structure Construction for Segmenting Industrial CT Volumes
Sho Watanabe, Yutaka Ohtake, Yukie NAGAI, Hiromasa Suzuki · Sekkei Kougaku, Shisutemu Bumon Kouenkai kouen rombunshuu/Sekkei Kogaku, Shisutemu Bumon Koenkai koen ronbunshu · 2016
Computed tomography (CT) has attracted attention in the industrial manufacturing for the inspection and analysis of assembled parts owing to its ability to examine the interior of objects without damaging them. To detect deformation or gaps in assembled parts, the precise shape of each part must be extracted from CT volumes. Although there are many conventional image-processing techniques for classifying images into regions, e.g. graph-cuts, mean-shift and active counters, it is difficult to divide CT volumes into components and achieve sub-voxel precision with them. To address this problem, we proposed a segmentation and shape extraction system based on a Morse complex that achieves segmentation of multiple parts at high precision by building a hierarchical data structure from a CT volume. First, a hierarchical data structure which records the history of the recursively merged clusters of CT volumes is built following an algorithm of the topological persistence and water-drop method. Then, voxel-level boundary surfaces are created by the manual operation with a GUI software. Finally, polygon meshes of all the components are generated at the sub-voxel-level accuracy. We evaluated the efficiency and precision of this approach, relative to how much effort was required to complete the segmentation of CT volumes and how close the created meshes are to the original shapes, respectively.