Comparison of simulated annealing algorithms for image restoration

This paper presents a comparative study of four optimisation algorithms based on simulated annealing: the Gibbs Sampler, the Metropolis algorithm, the Iterated Conditional Modes, and an original method of random descent proposed by the authors. Comparison criteria that have been chosen are the convergence speed, the quality of the optimum obtained with a new specific energy function and the total computing time necessary for image restoration. The four algorithms are applied to the special case of restoration of images disturbed by a gaussian white noise. Moreover, a new function to be minimized and several efficient heuristics allowing to decrease computing time are proposed. Some numerical experiments are given.