• DocumentCode
    814108
  • Title

    Binary partitioning, perceptual grouping, and restoration with semidefinite programming

  • Author

    Keuchel, Jens ; Schno, Christoph ; Schellewald, Christian ; Cremers, Daniel

  • Author_Institution
    Dept. of Math. & Comput. Sci., Mannheim Univ., Germany
  • Volume
    25
  • Issue
    11
  • fYear
    2003
  • Firstpage
    1364
  • Lastpage
    1379
  • Abstract
    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.
  • Keywords
    combinatorial mathematics; convex programming; graph theory; image restoration; binary decision variables; binary partitioning; binary restoration; combinatorial problem; combinatorial solutions; computer vision; convex optimization; convex programming; figure-ground discrimination; ground-truth experiments; interior-point methods; linear constraints; metric pairwise interactions; objective criterion; optimization method; perceptual grouping; performance bounds; quadratic functional minimization; randomized hyperplane technique; semidefinite programming relaxations; spectral graph theory; symmetry condition; unsupervised partitioning; Computer vision; Constraint optimization; Design optimization; Functional programming; Graph theory; Linear programming; Markov random fields; Optimization methods; Pattern recognition; Quadratic programming;
  • fLanguage
    English
  • Journal_Title
    Pattern Analysis and Machine Intelligence, IEEE Transactions on
  • Publisher
    ieee
  • ISSN
    0162-8828
  • Type

    jour

  • DOI
    10.1109/TPAMI.2003.1240111
  • Filename
    1240111