Abstract

In the early 80's, Smith and Barrett [1] have introduced Simulated Annealing techniques (SA) in the domain of tomographic image reconstruction. Since, this technique has been tested in various domains of Medical Imaging [2]-[3]. The main advantage of SA is to theoritically converge to the global minimum of a cost or energy function. In reconstruction imaging, this cost function contains two terms : one term which represents the mean squared error between the measured and estimated data, and another term, which acts as regulazrization and translates a priori information. In another approach, Geman and Geman have showed that SA converges to the Maximum A Posteriori solution in a bayesian estimation [4].

© 1992 Optical Society of America