Information Theoretic Studies and Assessment of Space Object Identification

摘要

This document represents the final report on the various scientific activities and accomplishments relating to Grant No. FA9550-09-1-0495 over its period of performance, June 1, 2009 - November 30, 2013. Among the most important technical accomplishments of the project are: * A new relation between two seemingly different metrics of information - mutual information (MI) and estimation theoretic Fisher information (FI) in the high-SNR limit, for arbitrary input and channel probability distributions; * New numerically efficient, tight upper and lower bounds on successfully transmitted statistical information, namely MI, in the estimation of one or more features of a target based on image data; (1) Improved approaches for an exact numerical, Monte-Carlo- sampling based computation of MI and comparison with our tight upper bounds; (2) Bayesian approaches to the problem of optimal selection of regularizer and its strength based on data themselves; (3) A full Bayesian error analyses of the prospects of 3D localization and superresolution of unresolved sources beyond the diffraction limit.