Enhancing digital images for maximum interpretability using linear programming

The feasibility of a new 1 digital image enhancement technique is described. The subject technique involves four hypotheses: (1) Image interpretability cau be related to objective quantitative figures of merit. (2) The target distributions that maximize these figures of merit can be digitally computed from the image and the point spread function. (3) A practical method for performing these calculations is available from linear programming. (4) If the point spread function is unknown except to within some finite uncertainty, the point spread functions that maximize these figures of merit can be calculated. This paper addresses the second and third of these. The first is reported by others authors in the open literature [14]. The fourth remains as yet unstudied. An experiment was conducted ranging over three targets, two point spread functions, four figures of merit, and fourteen types of additive noise. Of the 336 possible combinations, 40 were processed using linear programming. Fourier reconstruction was performed as well for comparison. The results have been assessed using several image quality measures. Familiarity with linear programming is assumed.