On the problem of stable image restoration (Q1341311)

To obtain accurate information about an object that is contained in its blurred image is usually extremely difficult. For this reason only some functionals on the object can be recovered. It is natural to choose, as functionals, the principal components, that is the linear combinations of the parameter estimates which contain the main bulk of information about the object. In this paper the authors show that the principal components can be found by an orthogonal transform, based on the eigenvectors of the Fisher information matrix. As a result, the stable estimate of the object can be constructed by using only the inner resources of the theory, without referring to the Bayes way of consideration. An algorithm of image restoration is described which takes into account both the photon noise and non-negativity of the object, being valid for the space-variant point spread function.