Double Annealing Regimes in the Multiple Birth-and-Death Stochastic Algorithms
Abstract
We consider relaxation properties of new multiple birth and death (MBAD) stochastic algorithms developed recently for problems of detection and feature extraction in image analysis. The MBAD algorithms are based on birth-and-death stochastic dynamics in continuum and its discrete time ap- proximation. We found the relation between cooling parameters in the MBAD algorithms (temperature and discretization step) guaranteeing the convergence to a measure supported on configurations giving the global minima of the energy function.