Image restoration with minimum a priori information

A consistent approach to the image restoration problem is presented, wich does not use Bayesian a priori information. Photon noise is taken into account. The unknown object is treated as a multidimensional set of parameters that have to be statistically estimated in an efficient way. The approach is based on an extended notion of feasible estimate (in the sense of information theory) and on Occam's razor rule of choosing the simplest object which is consistent with the data. Occam's rule is applied by transformation to principal components of the inverse (or maximum likelihood) estimate, which are generated by Fisher's information matrix. The same approach can also be applied to various other inverse problems.