Skeleton-guided 3D Shape Distance Field Metamorphosis

Graphical Models(Proceedings of CVM 2016)

Bo Wu1,2,          Kai Xu1,2,          Yang Zhou2,          Yueshan Xiong1,           Hui Huang3,2

1HPCL, National University of Defense Technology,       2Shenzhen VisuCA Key Lab / SIAT,       3Shenzhen University


Figure 1: Morphing two different structural shapes automatically. Top row: morphing results of [Weng et al. 2013]. Bottom row: morphing sequence based on our algorithm. Note both results use the same anchors obtained by our method and there is no anchor on gecko’s two lower legs.


Abstract

We introduce an automatic 3D shape morphing method without the need of manually placed anchor correspondence points. Given a source and a target shape, our approach extracts their skeletons and computes the meaningful anchor points based on their skeleton node correspondences. Based on the anchors, dense correspondences between the interior of source and target shape can be established using earth movers distance (EMD) optimization. Skeleton node correspondence, estimated with a voting-based method, leads to part correspondence which can be used to confine the dense correspondence within matched part pairs, thus providing smooth and plausible morphing results based on distance field interpolation (DFI). We demonstrate our algorithm works well with experimental results, including shapes with large geometry variation and structure difference.


Results

Figure 2: Morphing hand to plant. Top row: morphing sequence generated using traditional EMD optimization. The correspondence drift problem affects much on the results due to large proportion difference. Bottom row: morphing sequence produced based on our part-based EMD optimization algorithm.

Figure 3: More morphing results generated by our method. Objects in top two rows have same genus 0 and can be morphed to their opponents automatically. The last two rows show morphing results with different genus, but correspondences of some skeleton branches need to be set up by user.


Acknowledgments

We thank all the reviewers for their comments and feedback. We would also like to acknowledge our research grants: NSFC (61572507, 61202333,61379103), 973 Program (2014CB360503), Guangdong Science and Technology Program (2015A030312015, 2014B050502009, 2014TX01X033), Hunan Innovation Program (CX2012B027), Shenzhen VisuCA Key Lab (CXB201104220029A).


Bibtex

@ARTICLE{3Dmorph2016,
    title = {Skeleton-guided 3D shape distance field metamorphosis},
    author = {Bo Wu, Kai Xu, Yang Zhou, Yueshan Xiong, Hui Huang},
    journal = {Graphical Models},
    volume = {85},
    pages = {37–45},
    year = {2016}


Downloads
Copyright © 2016-2018 Visual Computing Research Center