![](bg8.png)
formulation of HS, instead of its standard ML form.
To this end, we formulated and compared a depth-
based, a normal-based and a specially tailored depth-
normal consistency prior. We conclude that correctly
utilising the given of piece-wise surface smoothness
in the MAP formulation greatly improves both local
and global reconstruction accuracy relative to ML re-
sults. Both quantitative and qualitative results indi-
cate our depth-normal consistency prior to be the cor-
rect formulation of the smoothness term, which by
enforcing consistency between depth and normal in-
formation produces the best results in terms of both
local smoothness and global object shape. The re-
sults generated with the prior are uniquely consistent
in both depth and integrated normal domain with the
normals being indexed from a geometrically correct
depth map. The computational overhead for the prior
is dependent on the size of the sampled voxel volume
and for the presented real data ranges from 2 min-
utes (billiard ball) to 4 hours (teapot no. 2). In our
future work we are confident the run-times will be re-
duced substantially by embedding MRF optimisation
of Bayesian HS into a coarse-to-fine framework using
octrees and/or by parallelising and porting prior cost
pre-computation (the bottleneck of the pipeline) onto
the GPU. In addition, we shall explore the potential
of our MRF framework in resolving ambiguities in
the sensor saturation region and at grazing sampling
angles.
REFERENCES
Baumgard, B. (1974). Geometric Modeling for Computer
Vision. PhD thesis, University of Stanford.
Delaunoy, A., Prados, E., and Belhumeur, P. (2010). To-
wards full 3D Helmholtz stereovision algorithms. In
Proc. of ACCV, volume 1, pages 39–52.
Frankot, R. and Chellappa, R. (1988). A method for enforc-
ing integrability in shape from shading algorithms.
PAMI, 10(4):439–451.
Guillemaut, J.-Y., Drbohlav, O., Illingworth, J., and
ˇ
S
´
ara,
R. (2008). A maximum likelihood surface normal es-
timation algorithm for Helmholtz stereopsis. In Proc.
of VISAPP, volume 2, pages 352–359.
Guillemaut, J.-Y., Drbohlav, O.,
ˇ
S
´
ara, R., and Illingworth,
J. (2004). Helmholtz stereopsis on rough and strongly
textured surfaces. In Proc. of 3DPVT, pages 10–17.
Helmholtz, H. (1925). Treatise on Physiological Optics,
volume 1. Dover (New York).
Kazhdan, M., Bolitho, M., and Hoppe, H. (2006). Poisson
surface reconstruction. In Proc. of SGP, pages 61–70.
Kolmogorov, V. (2006). Convergent tree-reweighted
message passing for energy minimization. PAMI,
28(10):1568 – 1583.
Kolmogorov, V. and Zabih, R. (2004). What energy func-
tions can be minimized via graph cuts? PAMI, 26:65–
81.
Laurentini, A. (1994). The visual hull concept
for silhouette-based image understanding. PAMI,
16(2):150–162.
Li, S. (1994). Markov random field models in computer
vision. In Proc. of ECCV, volume B, pages 361–370.
POV-Ray (2013). POV-Ray - The Persistence of Vision
Raytracer. http://www.povray.org/.
Scharstein, D. and Szeliski, R. (2002). A taxonomy and
evaluation of dense two-frame stereo correspondence
algorithms. IJCV, 47(1-3):7–42.
Seitz, S., Curless, B., Diebel, J., Scharstein, D., and
Szeliski, R. (2006). A comparison and evaluation of
multi-view stereo reconstruction algorithms. In Proc.
of CVPR, volume 1, pages 519– 528.
Szeliski, R., Zabih, R., Scharstein, D., Veksler, O., Kol-
mogorov, V., Agarwala, A., Tappen, M., and Rother,
C. (2008). A comparative study of energy min-
imization methods for markov random fields with
smoothness-based priors. PAMI, 30(6):1068–1080.
Tu, P., Mendonc¸a, P. R., Ross, J., and Miller, J. (2003). Sur-
face registration with a helmholtz reciprocity image
pair. In Proc. of IEEE Workshop on Color and Photo-
metric Methods in Computer Vision.
Wainwright, M. J., Jaakkola, T. S., and Willsky,
A. S. (2005). Map estimation via agreement on
trees: Message-passing and linear-programming ap-
proaches. IEEE Transactions on Information Theory,
51(11):3697–3717.
Weinmann, M., Ruiters, R., Osep, A., Schwartz, C., and
Klein, R. (2012). Fusing structured light consistency
and helmholtz normals for 3D reconstruction. In Proc.
of BMVC, pages 108.1–108.12. BMVA Press.
Woodham, R. J. (1989). Shape from shading, chapter Pho-
tometric method for determining surface orientation
from multiple images, pages 513–531. MIT Press,
Cambridge, MA, USA.
Wu, T.-P., Tang, K.-L., Tang, C.-K., and Wong, T.-T.
(2006). Dense photometric stereo: A Markov Ran-
dom Field approach. PAMI, 28(11):1830–1846.
Zickler, T. (2006). Reciprocal image features for uncali-
brated Helmholtz stereopsis. In Proc. of CVPR, pages
1801– 1808.
Zickler, T., Belhumeur, P. N., and Kriegman, D. J. (2002).
Helmholtz stereopsis: Exploiting reciprocity for sur-
face reconstruction. IJCV, 49(2–3):215–227.
Zickler, T. E., Ho, J., Kriegman, D. J., Ponce, J., and Bel-
humeur, P. N. (2003). Binocular Helmholtz stereopsis.
In Proc. of ICCV, volume 2, pages 1411–1417.
VISAPP2014-InternationalConferenceonComputerVisionTheoryandApplications
342