In most applications of image deconvolution to physical sciences accurate
information about statistical properties of the noise is available. This
information leads in a natural way to deconvolution methods based on the
Maximum Likelihood (ML) approach. The purpose of this tutorial is to
introduce basic facts and ideas about image deconvolution, to discuss
properties of noise and show how they lead to ML-problems. The iterative
methods most frequently used for solving these problems in practice are
also given. Moreover, since ML-approaches in general lead to ill-posed
problems, regularization is needed. This can be obtained by early stopping
of the iterations (semi-convergence property). However, by taking into
account the statistical setting of ML-problems, a quite natural approach is
provided by the so-called Bayesian methods. These methods are briefly
introduced and relationships with classical regularization methods are
discussed.
Copyright. All Rights Reserved.