- 最后登录
- 2018-6-29
- 注册时间
- 2011-7-1
- 阅读权限
- 20
- 积分
- 359
- 纳金币
- 335582
- 精华
- 0
|
Multigrid and Multilevel Preconditioners for Computational Photography
Dilip Krishnan Richard Szeliski
Department of Computer Science Interactive Visual Media Group
New York University Microsoft Research
Abstract
This paper unifies multigrid and multilevel (hierarchical) precon-
ditioners, two widely-used approaches for solving computational
photography and other computer graphics simulation problems. It
provides detailed experimental comparisons of these techniques
and their variants, including an analysis of relative computational
costs and how these impact practical algorithm performance. We
derive both theoretical convergence rates based on the condition
numbers of the systems and their preconditioners, and empirical
convergence rates drawn from real-world problems. We also de-
velop new techniques for sparsifying higher connectivity problems,
and compare our techniques to existing and newly developed vari-
ants such as algebraic and combinatorial multigrid. Our experimen-
tal results demonstrate that, except for highly irregular problems,
adaptive hierarchical basis function preconditioners generally out-
perform alternative multigrid techniques, especially when compu-
tational complexity is taken into account.
Keywords: Computational photography, Poisson blending, col-
orization, multilevel techniques, fast PDE solution, parallel algo-
rithms
1 Introduction
Multigrid andmultilevel preconditioning techniques have long been
widely used in computer graphics and computational photography
as ameans of accelerating the solution of large gridded optimization
problems such as geometric modeling [Gortler and Cohen 1995],
high-dynamic range tone mapping [Fattal et al. 2002], Poisson and
gradient-domain blending [P´ erez et al. 2003; Levin et al. 2004b;
Agarwala et al. 2004], colorization [Levin et al. 2004a] (Fig. 1), and
natural image matting [Levin et al. 2008]. They have also found
widespread application in the solution of computer vision prob-
lems such as surface interpolation, stereo matching, optical flow,
and shape from shading [Terzopoulos 1986; Szeliski 1990; Pent-
land 1994], as well as large-scale finite element and finite difference
modeling [Briggs et al. 2000; Trottenberg et al. 2000].
While the locally adaptive hierarchical basis function technique de-
veloped by Szeliski [Szeliski 2006] showed impressive speedups
over earlier non-adaptive basis functions [Szeliski 1990], it was
never adequately compared to state-of-the art multigrid techniques
such as algebraic multigrid [Briggs et al. 2000; Trottenberg et al.
2000] or to newer techniques such as combinatorial multigrid
[Koutis et al. 2009]. Furthermore, the original technique was re-
stricted to problems defined on four neighbor (N4) grids.
In this paper, we generalize the sparsification method introduced in
[Szeliski 2006] to handle a larger class of grid topologies, and show
how multi-level preconditioners can be enhanced with smoothing to
create hybrid algorithms that accrue the advantages of both adap-
tive basis preconditioning and multigrid relaxation. We also pro-
vide a detailed study of the convergence properties of all of these
algorithms using both condition number analysis and empirical ob-
servations of convergence rates on real-world problems in computer
graphics and computational photography.
全文请下载附件:
|
|