Multiview Reconstruction of Complex Organic Shapes

Jasenko Zivanov and Thomas Vetter

Abstract

We propose a novel narrow baseline multiview stereo surface reconstruction method that is specifically aimed at complex shapes of biological origin that show many thin protrusions and curved occluding contours. Our method is built around fitting local quadrics to occluding contours and it thus avoids any planarity assumptions that are common to other state of the art methods. We describe a complete pipeline that begins with calibrated noisy images and produces a final watertight surface. We present a novel technique to detect pixel precise internal contours and to fit local quadrics to them. This procedure is designed to deal with curved occluding contours, and it is very robust to noise and to slow changes in surface radiance. Our method can even reconstruct shapes from sequences where the illumination is attached to the observer and not the scene. We demonstrate the potential of our method by reconstructing the intricate shape of a tiny insect from images taken under a scanning electron microscope.

Session

Poster 2

Files

Extended Abstract (PDF, 1788K)
Paper (PDF, 3M)

DOI

10.5244/C.29.157
https://dx.doi.org/10.5244/C.29.157

Citation

Bibtex

@inproceedings{BMVC2015_157,
	title={Multiview Reconstruction of Complex Organic Shapes},
	author={Jasenko Zivanov and Thomas Vetter},
	year={2015},
	month={September},
	pages={157.1-157.11},
	articleno={157},
	numpages={11},
	booktitle={Proceedings of the British Machine Vision Conference (BMVC)},
	publisher={BMVA Press},
	editor={Xianghua Xie, Mark W. Jones, and Gary K. L. Tam},
	doi={10.5244/C.29.157},
	isbn={1-901725-53-7},
	url={https://dx.doi.org/10.5244/C.29.157}
}