An Algebraic Approach to Surface Reconstruction from Gradient Fields
Amit Agrawal, Rama Chellappa
Center for Automation Research
University of Maryland
College Park, MD 20742 USA
(aagrawal,rama)@cfar.umd.edu
Ramesh Raskar
Mitsubishi Electric Research Labs
201 Broadway
Cambridge, MA 02139 USA
raskar@merl.com
Abstract
Several important problems in computer vision such as
Shape from Shading (SFS) and Photometric Stereo (PS) re-
quire reconstructing a surface from an estimated gradient
field, which is usually non-integrable, i.e. have non-zero
curl. We propose a purely algebraic approach to enforce
integrability in discrete domain. We first show that enforc-
ing integrability can be formulated as solving a single lin-
ear system Ax = b over the image. In general, this system
is under-determined. We show conditions under which the
system can be solved and a method to get to those condi-
tions based on graph theory. The proposed approach is
non-iterative, has the important property of local error con-
finement and can be applied to several problems. Results on
SFS and PS demonstrate the applicability of our method.
1. Introduction
The notion of integrability arises whenever a surface has
to be reconstructed from a gradient field. In several core
computer vision problems such as Shape from Shading and
Photometric Stereo, an estimate of the gradient field is avail-
able. The gradient field is then integrated to obtain the de-
sired 2D surface (shape). However, the estimated gradient
field often has non-zero curl making it non-integrable. In
this paper, we address the problem due to curl and present
a method to enforce integrability. The approach is non-
iterative and has the important property that the errors due
to non-zero curl do not propagate across the image.
We present an algebraic approach where integrability is
enforced by finding a residual gradient field. Adding the
computed residual gradients to the specified gradients pro-
duces an integrable gradient field. Our main contributions
are as follows:
• We present a method to exploit the information con-
tained in the curl of the given non-integrable field. In
discrete domain, the residual gradient field and the curl
form a linear system which we use to achieve a better
reconstruction.
• When curl is non-zero at several pixels, the linear
system is potentially under-determined with more un-
knowns (residual gradient values) than the number of
equations (corresponding to known curl values). Us-
ing a graph analogy, we derive conditions under which
the residual field can be recovered and show a method
to achieve those conditions.
• Unlike least square approaches, our approach has the
property of local error confinement and the err or in
reconstructed surface does not spread throughout the
gradient field. Due to its global nature, our method
is non-iterative compared to iterative techniques [9]
which may have convergence issues.
1.1. Related work
Researchers have addressed the issue of enforcing in-
tegrability typically specific to the problem at hand. In
Shape from Shading algorithms such as [17][8], integra-
bility was enforced as a constraint in the minimization rou-
tine. Frankot & Chellappa [6] enforce integrability by or-
thogonally projecting the non-integrable field on to a vector
subspace spanning the set of integrable slopes. However,
their method is dependent on the choice of basis functions.
Simchony et. al. [11] find the integrable gradient field clos-
est to the given gradient field in a least squares sense by
solving the Poisson equation. One can show that (see sec-
tion 2) their method ignores the information in the curl and
finds a zero-curl field which has the same divergence as the
given non-integrable field. The method also lacks the prop-
erty of error confinement. For a survey of SFS algorithms
see [16].
Photometric s tereo [13][5] uses multiple images ob-
tained under different illumination directions to recover the
surface gradients. In [9], belief propagation in graphical
networks was used to enforce integrability for SFS and PS
problems. In [14], the integrability constraint was used to