Partial Symmetry Detection in Volume Data

Abstract

In this paper, we present an algorithm for detecting partial Euclidean symmetries in volume data. Our algorithm finds subsets in voxel data that map to each other approximately under translations, rotations, and reflections. We implement the search for partial symmetries efficiently and robustly using a feature-based approach: We first reduce the volume to salient line features and then create transformation candidates from matching only local configurations of these line networks. Afterwards, only a shortlist of transformation candidates need to be verified using expensive dense volume matching. We apply our technique on both synthetic test scenes as well as real CT scans and show that we can recover a large amount of partial symmetries for complexly structured volume data sets.