Maximum Entropy Image Reconstruction

Abstract

Two-dimensional digital image reconstruction is an important imaging process in many of the physical sciences. If the data are insufficient to specify a unique reconstruction, an additional criterion must be introduced, either implicitly or explicitly before the best estimate can be computed. Here we use a principle of maximum entropy, which has proven useful in other contexts, to design a procedure for reconstruction from noisy measurements. Implementation is described in detail for the Fourier synthesis problem of radio astronomy. The method is iterative and hence more costly than direct techniques; however, a number of comparative examples indicate that a significant improvement in image quality and resolution is possible with only a few iterations. A major component of the computational burden of the maximum entropy procedure is shown to be a two-dimensional convolution sum, which can be efficiently calculated by fast Fourier transform techniques.