Optimisation Algorithms in Probabilistic Relaxation Labelling
The use of optimisation approaches for relaxation labelling is reviewed and its relationship to earlier heuristic schemes is considered. The fixed points of optimisation schemes are determined by the constraint relationships and for simple examples can be predicted. Optimisation techniques ensure faster convergence of the relaxation process and can incorporate a wider class of constraints than the heuristic methods. These properties are illustrated using simple “toy” problems.
KeywordsDescent Direction Steep Descent Method Labelling Problem Probability Component Optimal Step Size
Unable to display preview. Download preview PDF.
- Kittler J. and Hancock E.R., “A contextual decision rule based on triplets of object-labels”, Alvey Vision Club Conference, Bristol, U.K. September 1986.Google Scholar
- O’Leary D.P. and Peleg S., “Analysis of relaxation processes: the two node — two label case”, IEEE SMC-13, No 4, pp 618–623 (1983)Google Scholar
- Ostrowski, Solutions of Equations and Systems of Equations, Academic Press, New YorkGoogle Scholar