Binary partitioning, perceptual grouping, and restoration with semidefinite programming

被引:65
作者
Keuchel, J [1 ]
Schnörr, C [1 ]
Schellewald, C [1 ]
Cremers, D [1 ]
机构
[1] Univ Mannheim, Dept Math & Comp Sci, CVGPR Grp, D-38131 Mannheim, Germany
关键词
image partitioning; segmentation; graph cuts; perceptual grouping; figure-ground discrimination; combinatorial optimization; relaxation; convex optimization; convex programming;
D O I
10.1109/TPAMI.2003.1240111
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We introduce a novel optimization method based on semidefinite programming relaxations to the field of computer vision and apply it to the combinatorial problem of minimizing quadratic functionals in binary decision variables subject to linear constraints. The approach is (tuning) parameter-free and computes high-quality combinatorial solutions using interior-point methods (convex programming) and a randomized hyperplane technique. Apart from a symmetry condition, no assumptions (such as metric pairwise interactions) are made with respect to the objective criterion. As a consequence, the approach can be applied to a wide range of problems. Applications to unsupervised partitioning, figure-ground discrimination, and binary restoration are presented along with extensive ground-truth experiments. From the viewpoint of relaxation of the underlying combinatorial problem, we show the superiority of our approach to relaxations based on spectral graph theory and prove performance bounds.
引用
收藏
页码:1364 / 1379
页数:16
相关论文
共 70 条
[1]   A generic grouping algorithm and its quantitative analysis [J].
Amir, A ;
Lindenbaum, M .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 1998, 20 (02) :168-185
[2]  
[Anonymous], 1982, ESTIMATION DEPENDENC
[3]  
[Anonymous], 1997, QUALITY SEMIDEFINITE
[4]   PROBLEMS OF DISTANCE GEOMETRY AND CONVEX PROPERTIES OF QUADRATIC MAPS [J].
BARVINOK, AI .
DISCRETE & COMPUTATIONAL GEOMETRY, 1995, 13 (02) :189-202
[5]   Mixed linear and semidefinite programming for combinatorial and quadratic optimization [J].
Benson, SJ ;
Ye, YY ;
Zhang, X .
OPTIMIZATION METHODS & SOFTWARE, 1999, 11-2 (1-4) :515-544
[6]   ILL-POSED PROBLEMS IN EARLY VISION [J].
BERTERO, M ;
POGGIO, TA ;
TORRE, V .
PROCEEDINGS OF THE IEEE, 1988, 76 (08) :869-889
[7]  
BESAG J, 1986, J R STAT SOC B, V48, P259
[8]  
Blake A., 1987, Visual Reconstruction
[9]  
Boppana R, 1987, P 28 IEEE S FDN COMP, P280
[10]   On the estimation of Markov random field parameters [J].
Borges, CF .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 1999, 21 (03) :216-224