Point-cloud registration using adaptive radial basis functions

Abstract

Non-rigid registration is a common part of bioengineering model-generation workflows. Compared to common mesh-based methods, radial basis functions can provide more flexible deformation fields due to their meshless nature. We introduce an implementation of RBF non-rigid registration with iterative knot-placement to adaptively reduce registration error. The implementation is validated on surface meshes of the femur, hemi-pelvis, mandible, and lumbar spine. Mean registration surface errors ranged from 0.37 to 0.99 mm, Hausdorff distance from 1.84 to 2.47 mm, and DICE coefficients from 0.97 to 0.99. The implementation is available for use in the free and open-source GIAS2 library.

Keywords:

• Non-rigid registration
• registration
• radial basis function
• morphing
• model generation

Notes

Acknowledgements

The authors would like to acknowledge the Melbourne Femur Collection and the Victorian Institute of Forensic Medicine for providing the CT scans and segmentations used in this work.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 https://bitbucket.org/jangle/gias2

2 http://gias2-shape-modelling-tutorial.readthedocs.io/en/latest/