Generalizing Swendsen-Wang to sampling arbitrary posterior probabilities

被引:103
|
作者
Barbu, A [1 ]
Zhu, SC [1 ]
机构
[1] Univ Calif Los Angeles, Dept Comp Sci & Stat, Los Angeles, CA 90095 USA
基金
美国国家科学基金会;
关键词
Swendsen-Wang; cluster sampling; Markov chain Monte Carlo; Bayesian inference; image segmentation; stereo matching;
D O I
10.1109/TPAMI.2005.161
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Many vision tasks can be formulated as graph partition problems that minimize energy functions. For such problems, the Gibbs sampler [9] provides a general solution but is very slow, while other methods, such as Ncut [24] and graph cuts [4], [22], are computationally effective but only work for specific energy forms [17] and are not generally applicable. In this paper, we present a new inference algorithm that generalizes the Swendsen-Wang method [25] to arbitrary probabilities defined on graph partitions. We begin by computing graph edge weights, based on local image features. Then, the algorithm iterates two steps. 1) Graph clustering: It forms connected components by cutting the edges probabilistically based on their weights. 2) Graph relabeling: It selects one connected component and flips probabilistically, the coloring of all vertices in the component simultaneously. Thus, it realizes the split, merge, and regrouping of a "chunk" of the graph, in contrast to Gibbs sampler that flips a single vertex. We prove that this algorithm simulates ergodic and reversible Markov chain jumps in the space of graph partitions and is applicable to arbitrary posterior probabilities or energy functions defined on graphs. We demonstrate the algorithm on two typical problems in computer vision-image segmentation and stereo vision. Experimentally, we show that it is 100-400 times faster in CPU time than the classical Gibbs sampler and 20-40 times faster then the DDMCMC segmentation algorithm [27]. For stereo, we compare performance with graph cuts and belief propagation. We also show that our algorithm can automatically infer generative models and obtain satisfactory results ( better than the graphic cuts or belief propagation) in the same amount of time.
引用
收藏
页码:1239 / 1253
页数:15
相关论文
共 50 条
  • [21] PARALLELIZATION OF THE 2D SWENDSEN-WANG ALGORITHM
    HACKL, R
    MATUTTIS, HG
    SINGER, JM
    HUSSLEIN, T
    MORGENSTERN, I
    INTERNATIONAL JOURNAL OF MODERN PHYSICS C-PHYSICS AND COMPUTERS, 1993, 4 (06): : 1117 - 1130
  • [22] Mixing properties of the Swendsen-Wang process an classes of graphs
    Cooper, C
    Frieze, AM
    RANDOM STRUCTURES & ALGORITHMS, 1999, 15 (3-4) : 242 - 261
  • [23] Rapid mixing of Swendsen-Wang dynamics in two dimensions
    Ullrich, Mario
    DISSERTATIONES MATHEMATICAE, 2014, (502) : 5 - +
  • [24] MEAN-FIELD STUDY OF THE SWENDSEN-WANG DYNAMICS
    RAY, TS
    TAMAYO, P
    KLEIN, W
    PHYSICAL REVIEW A, 1989, 39 (11): : 5949 - 5953
  • [25] Swendsen-Wang dynamics for the ferromagnetic Ising model with external fields
    Feng, Weiming
    Guo, Heng
    Wang, Jiaheng
    INFORMATION AND COMPUTATION, 2023, 294
  • [26] Swendsen-Wang dynamics for general graphs in the tree uniqueness region
    Blanca, Antonio
    Chen, Zongchen
    Vigoda, Eric
    RANDOM STRUCTURES & ALGORITHMS, 2020, 56 (02) : 373 - 400
  • [27] SWENDSEN-WANG IS FASTER THAN SINGLE-BOND DYNAMICS
    Ullrich, Mario
    SIAM JOURNAL ON DISCRETE MATHEMATICS, 2014, 28 (01) : 37 - 48
  • [28] Swendsen-Wang algorithm on the mean-field Potts model
    Galanis, Andreas
    Stefankovic, Daniel
    Vigoda, Eric
    RANDOM STRUCTURES & ALGORITHMS, 2019, 54 (01) : 82 - 147
  • [29] EFFICIENT PARALLELIZATION OF THE 2D SWENDSEN-WANG ALGORITHM
    HACKL, R
    MATUTTIS, HG
    SINGER, JM
    HUSSLEIN, T
    MORGENSTERN, I
    PHYSICA A, 1994, 212 (3-4): : 261 - 276
  • [30] RIGOROUS LOWER BOUND ON THE DYNAMIC CRITICAL EXPONENTS OF THE SWENDSEN-WANG ALGORITHM
    LI, XJ
    SOKAL, AD
    PHYSICAL REVIEW LETTERS, 1989, 63 (08) : 827 - 830