Stochastic Optimization Algorithms: An Application to Pattern Matching

摘要

The authors have investigated the Langevin stochastic equation of motion with application to a certain optimization problem - pattern matching in digital images. The method is similar to simulated annealing techniques with the exception that here the equations of motion govern the path in parameter space, where the number of parameters equal the degrees of freedom in the optimization. A brief derivation of the stationary Boltzmann distribution is made. A model with inertia is also described. The application to multiparameter contour matching is investigated and the benefits of the model are shown.