Stochastic learning automata for image estimation

Maximum posterior marginals (MPM) image estimation is examined in the framework of stochastic learning automata. An automaton is associated with each pixel, with the M gray levels possible at the pixel constituting the action set of the automaton. This distributed set of automata function in a cooperative manner with each automaton using an identical and predefined stochastic learning scheme to converge in probability to its optimal action. Three different learning schemes are studied and a parallel implementation of each learning scheme is given. Experimental results of reconstructing the original from a noisy image are presented.<>