MAP estimation in image restoration by a local search enhanced genetic algorithm

Investigates a maximum a posteriori distribution (MAP) estimation approach to image restoration. Images are modelled as Markov random fields (MRF), a form of 2-D stochastic process. The authors present a new MAP estimation algorithm which combines a global search method, the genetic algorithm (GA), and a fast local search strategy, iterated conditional modes (ICM). This hybrid algorithm combines the fast convergence of ICM with the power of sustained exploration from the GA in order to achieve a good optimum, if not the global one. the major modification to the GA is the introduction of local hill climbing and voting controlled mutation mechanisms. The algorithm is used to restore two artificial images, both distorted by an independent additive Gaussian noise. The results are compared with that from ICM alone, and simulated annealing (SA). It is shown that the new MAP algorithm reaches a good optimum in a significantly smaller number of iterations compared with SA