scientific article; zbMATH DE number 4001287

From MaRDI portal
Publication:4727261





[For the entire collection see Zbl 0597.00025.] We address the problem of reconstructing an image from incomplete and noisy data. Maximum entropy is to be preferred to the direct Bayesian approach, which leads to impossibly large problems of computation and display in the space of all images. The monkey approach to maximum entropy, in which N balls are thrown randomly at the cells of an image, invites us to interpret entropy as the logarithm of a prior probability. Such an interpretation leads to severe difficulties with the parameter N, as well as leaving unsolved the problem of selecting a single image for final display. It is better to justify maximum entropy by the kangaroo argument. This shows that maximum entropy is the only generally applicable technique which will give the correct answer to certain simple problems. As originally presented, maximum entropy fails to incorporate many types of prior knowledge. This defect can be remedied by a small change of interpretation, with great potential benefit for the quality of reconstructed images.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4727261)