Maximum entropy image reconstruction - A practical non-information-theoretic approach

Abstract

An alternative motivation for the maximum entropy method (MEM) is given and its practical implementation discussed. The need for nonlinear restoration methods in general is considered, arguing in favor of nonclassical techniques such as MEM. Earlier work on MEM is summarized and the present approach is introduced. The whole family of restoration methods based on maximizing the integral of some function of the brightness is addressed. Criteria for the choice of the function are given and their properties are discussed. A parameter for measuring the resolution of the restored map is identified, and a scheme for controlling it by adding a constant to the zero-spacing correlation is introduced. Numerical schemes for implementing MEM are discussed and restorations obtained with various choices of the brightness function are compared. Data noise is discussed, showing that the standard least squares approach leads to a bias in the restoration.