Maximum entropy image processing using transform domain constraints

A formulation of the maximum entropy (ME) method is described, where the data constraints are expressed in the form of fixed bounds on the elements of an orthogonal transform of the model. The bounds are set on the basis of both the observed data and an estimate of the noise statistics in the transform domain; prior knowledge, if available, can also be incorporated. Using a special-purpose conjugate gradient algorithm developed for this problem, one-dimensional examples are presented that illustrate substantial SNR enhancement using the new formulation with both Fourier and Walsh transforms. A simple strategy for selecting an initial feasible solution for the algorithm is presented.<>